Dawn Song - Adversarial Machine Learning and Computer Security _ Lex Fridman Podcast #95

Growth and its Relationship to Experience

Growth is often associated with experience, but they are not exactly the same thing. Growth can be thought of as a process of improvement and development, where an individual strives to become better versions of themselves every day. It's about trying new things, taking risks, and pushing oneself beyond what is comfortable or familiar. Experience, on the other hand, refers to the actual events and circumstances that shape our lives. While experience can be a catalyst for growth, it's not the same thing as growing.

Ultimately, growth is a state of mind. It's about embracing the journey, rather than just focusing on the destination. When we're focused solely on achieving a goal or outcome, we can become stuck in a mindset of achievement, where we're constantly striving for more. But what if the goal itself becomes the problem? What if the pursuit of success and happiness is actually the obstacle to finding true fulfillment?

The Evolution of Growth

Growth is not just about individual development; it's also deeply connected to the evolution of the world around us. The natural world is constantly changing, evolving, and adapting to new circumstances. This process of growth and transformation is mirrored in human experience, where we too must navigate our own growth and development.

The idea that growth is a necessary component of life is almost laughable when you consider the concept of status. We often associate success with external validation, like social media likes or professional recognition. But what if these external markers are actually the opposite of true fulfillment? What if the pursuit of status and recognition is just a distraction from the real work of growing and developing as individuals?

Finding Meaning in Growth

For people who dedicate themselves to finding answers to life's big questions, growth can be a powerful tool for discovery. By focusing on personal development and self-improvement, we can gain clarity on what truly matters to us. But here's the thing: growth is not just about achieving a specific goal or outcome; it's also about embracing the process itself.

The search for meaning in life is ultimately a subjective experience. We each must find our own way to define purpose and fulfillment. And yet, there's something liberating about knowing that we have the power to create our own meaning. When we focus on growth and development, we open ourselves up to new possibilities and opportunities.

The Question of Meaning

At its core, the question of meaning is a question of why anything matters at all. Why do we exist? What gives our lives significance? These are profound questions that have puzzled philosophers and thinkers for centuries. But what if the answer lies not in some grand, cosmic truth, but rather in our individual experiences and choices?

The Trap of Over-Questioning

Asking the question "what is the meaning of life?" can become a trap, leading us down a path of over-analysis and self-doubt. When we become too fixated on finding an answer, we risk losing sight of what truly matters: our own experiences, relationships, and contributions to the world.

Personal Story

I'll never forget when I shifted my focus from security to pursuing my true passion – building hydrogen machines. It was a turning point in my life, one that led me to become one of the top researchers in my field. In this moment, I realized that growth is not just about achieving external validation; it's also about following our inner voice and taking risks.

The Power of Focusing

When we focus on personal growth and development, we open ourselves up to new possibilities and opportunities. We begin to see the world in a different light, as a place where anything can happen if we're willing to take the leap. By letting go of external validation and embracing our own inner voice, we can find true fulfillment and purpose.

In Conclusion

Hacking is not just about solving problems or achieving success; it's also about playing with others, pushing boundaries, and exploring new possibilities. As Steve Wozniak once said, "A lot of hacking is playing with other people." This mindset of experimentation and exploration is essential for personal growth and development.

Final Thoughts

And so, we come to the end of this conversation on the meaning of life. I hope that our discussion has provided you with a deeper understanding of the complexities surrounding this question. Remember, growth is not just about achieving external validation; it's also about embracing the process itself. By focusing on personal development and self-improvement, we can gain clarity on what truly matters to us. And always remember, the search for meaning in life is ultimately a subjective experience – one that each of us must find our own way to define.

"WEBVTTKind: captionsLanguage: enthe following is a conversation with Dan song a professor of computer science at UC Berkeley with research interests and computer security most recently with a focus on the intersection between security and machine learning this conversation was recorded before the outbreak of the pandemic for everyone feeling the medical psychological and financial burden of this crisis I'm sending love your way stay strong we're in this together we'll beat this thing this is the artificial intelligence podcast if you enjoy it subscribe on YouTube review it with five stars on Apple podcast supported on patreon or simply connect with me on Twitter Alex Friedman spelled the Fri D M a.m. as usual I'll do a few minutes of ads now and never any ads in the middle that can break the flow of the conversation I hope that works for you it doesn't hurt the listening experience this show is presented by cash app the number one finance app in the App Store when you get it use collects podcast cash app lets you send money to friends buy Bitcoin and invest in the stock market with as little as $1 since cash app does fractional share trading let me mention that the order execution algorithm that works behind the scenes to create the abstraction of fractional orders is an algorithmic marvel so big props the cash app engineers for solving a hard problem that in the end provides an easy interface that takes a step up to the next layer of abstraction over the stock market making trading more accessible for new investors and diversification much easier so again if you get cash app from the App Store Google Play and use the code lex podcast you get ten dollars in cash wrap will also donate ten dollars the first an organization that is helping to advanced robotics and STEM education for young people around the world and now here's my conversation with dawn song systems will always have security vulnerabilities I started abroad almost philosophical level that's a very good question I mean in general right it's very difficult to write completely bug-free code and code that has no one in policy and also especially given that's the definition for nobility is actually really proud it's any type of attacks essentially an ax code can you know that's can you can cut out the cost by vulnerabilities and the nature of attacks is always changing as well like new parts are coming up okay so for example in the past we talked about memory safety type of vulnerabilities we're essentially tackers can exploit and the software and the take over control for how the code runs and then can launch attacks that way by accessing some aspect of the memory and be able to then alter the state of the program excite so for example in the example for buffer overflow then the attacker essentially actually causes essentially unintended changes in the states of the after program and then for example can then take over control flow after program and that the program to execute code that's actually the the programming design intent so the attack can be a remote attack so they the attacker for example can can send in a malicious input to the program that just causes a program to completely then be compromised and then end up doing something that's under the program and the attackers control and intention but that's just one form of attacks and there are other forms of attacks like for example there are these side channels where attackers can try to learn from even just observing the outputs from the behaviors of the program try to infer certain secrets of the program so they essentially write the form of attacks it's very very it's very broad spectrum and in general from the security perspective we want to essentially provide as much guarantee as possible about the program's security properties and so on so for example we talked about the provable guarantees of the program so for example there are ways we can use program analysis and form verification techniques to prove that a piece of code has no memory safety vulnerabilities what does that look like what does that proof is that just a dream for that's applicable to small case examples is that possible to do two for real-world systems so actually I mean today I actually call it so we are entering the area of formally verified systems so in the community we have been working for the past decades in developing techniques and tools to do this type of program verification and and we have dedicated teams that have dedicated you know they're like years sometimes even decades of their work in the space so as a result so we actually have a number of formally verify systems ranging from micro kernels to compilers to file systems to certain crypto you know libraries and so on and so it's actually really wide ranging and it's really exciting to see that people are recognizing the importance of having these formally verified systems with verified security so that's great advancement that we see but on the other hand I think we do need to take all these in essentially with with the culture as well in the sense that's just like I said the the type of vulnerability is very varied so we can form a very fine a software system to have certain set of security properties but they can still be vulnerable to other types of attacks and hence it's that we continue to make progress in the in the space so just a quick to linger on the formal verification is that something you can do by looking at the code alone or is it something you have to run the code to to prove something so empirical verification can you look at the code just the code so that's a very very question so in general for most program verification techniques is essentially try to verify the properties of the program statically and there are reasons for that too we can run the code to see for example using like in suffer testing with fasting techniques and also in certain even model checking techniques you can actually run the code but in general that only allows you to essentially verify or analyze the behaviors after program in certain and the certain situations and so most of the program verification techniques actually works statically what astatically mean that's the running the code without writing the code yep so what sort of to return this is the big question if we can stand that for a little bit longer do you think there will always be security vulnerabilities you know that's such a huge worry for people in the broad cyber security threat in the world it seems like the the tension between nations between groups the the Wars of the future might be fought in cyber security security that people worry about and so of course the nervousness is is this something that we can get a hold of in the future for our software systems so there's a very funny quotes seeing security is job security we strive to make progress in building more secure systems and also making it easier and easier to build secure systems but given and the diversity the the various nature of attacks and also the interesting thing about security is that unlike in most other views essentially we are trying to hash applets improve a statement true but in this case yes trying to say that there is no attacks so even just this demon itself it's not very well defined again given you know how vary the nature of the attacks can be it has there's a challenge of security and also then naturally essentially it's almost impossible to say that something a real-world system is a hundred percent no security vulnerabilities is there a particular and we'll talk about different kinds of vulnerabilities it's exciting ones very fascinating ones in the space of machine learning but is there a particular security vulnerability that worries you the most that you think about the most in terms of it being a really hard problem and a really important problem to solve so I have in the past have worked essentially through the Oh through the different stacks in the systems and I can networking security software security and even in social security there is our time program binary security and then web security mobile security so so throughout we have been developing more techniques and tools to improve security of the software systems and as a consequence actually is a very interesting thing that we are seeing an interesting trends that we're seeing is that the attacks are actually moving more anymore from the systems south yeah towards to humans so it's moving up the stack it's moving up the stack as faster and also it's moving more and more towards what we call the weakest link so we say though in security we say the weakest link actually have the system's oftentimes is actually humans themselves so a lot of attacks for example that hackers others through social engineering from these other methods they actually attack the humans and then attack the systems so we'll actually have projects that actually works on how to use a machine learning to help humans to defend against this effort actually so yeah so if we look at humans as security vulnerabilities is there is there methods is that what you're kind of referring to is there hope or methodology for pad the humans I think in the future this is going to be really mind more of a serious issue because again for for machines for systems we can yes we can patch them we can build a more secure systems we can harden them and so on but humans are actually we don't have a way to say to a software upgrade out to a hardware for humans and so for example right now we you know we already see different types of attacks in particularly I think in the future they are going to be even more effective on humans so as I mentioned social engineering attacks like these phishing attacks attackers I'll just get humans to provide their passwords and there have been instances where even places like Google and other places and that's supposed to have really good security people there have been fished to actually wire money to attackers and also we talked about this the fake and fake news so these essentially are there to target humans to manipulate humans opinions perceptions and so on and so I think in going to the future these are going to become more and more severe is further of the stack yes yes so so you see kind of social engineering automated social engineering as a kind of security vulnerability oh absolutely and again given that the humans are the weakest link to the system I I would say this is a type of attacks that I would be most worried about all that's fascinating okay so also we need to a I to help humans to as I mentioned we have some projects in the space actually helps and that can you maybe can go there for what are some ideas projects we are working on is actually using NLP and chat bot techniques to help humans for example the Chabad actually could be they're observing the conversation between a user and a remote pundants and then the checkout could be there to try to observe to see whether the correspondence is potentially attacker for example in some of the phishing attacks the attacker claims to be a relative of the user and the and the relative got lost in London and he's you know walleyes have been stolen had no money as the user to wire money to send money to the attacker right to the correspondent and so then in this case the Chabad actually could try to recognize and there may be some things the species going on and this relates to asking money to be sent and also the chibok could actually post and we call it challenge and response the correspondence claims to be a relative of the user then the checkout could automatically actually generate some kind of challenges to see whether the correspondence knows the appropriate knowledge to prove that he actually else he or she actually is the claimed in the relative after user so in the future I think these type of technologies actually could help protect users that's funny so get the so chat but that's kind of focused for looking for the kind of patterns that are usually usually associated with social engineering attacks right it would be able to then test sort of do a basic capture type of a response to see is this is the faction of the semantics of the claims you're making true right develop you know more powerful and now P and T bar techniques the chapel could even engage further conversations with the correspondence to for example if it turns out to be a and you know attack then the the the topic can try to engage in conversations with the attacker to try to learn more information from the attacker as well so it's a very interesting area so that chap I is essentially your your little representative in the spate in the security space it's like your little lawyer that protects you from doing anything stupid that's a fascinating vision for the future do you see that broadly applicable across the web so you across all your interactions what about like on social networks for example so across all of that do you see that being implemented in sort of that's the service that a company would provide or does every single social network has to implement it themselves so Facebook and Twitter and so on or do you see there being like a security service that kind of is a plug-and-play that's a very good question I think of course we still have a ways to go until the analogy and the tapout techniques can be that effective but I think it right once it's powerful enough I do see that that can be a service as a user can employ or can be deployed by the platforms it's just the curious side to me on security and we'll talk about privacy is who gets a little bit more of the control who gets to you know on whose side is the representative is it on Facebook side that there is this security protector or is it on your side and it has different implications about how much that little chatbot security protector knows about you nice exactly if you have a little security bot that you carry with you everywhere from Facebook to Twitter to all your services they might it might know a lot more about you and a lot more about your relatives to be able to test those things but that's okay because you have more control of that as opposed to Facebook having that that's a really interesting trade-off another fascinating topic you work on is again also non-traditional to think about a security vulnerability but I guess it is is adversarial machine learning is basically again high up the stack being able to attack the the accuracy the performance of this of machine learning systems by manipulating some aspect perhaps actually can clarify but I guess the traditional way the main way is to manipulate some the input data to make the output something totally not representative of the semantic content of the right so in this adversarial machine essentially attackers the goal is to fold the machining system me into making the wrong decision and the attack can actually happen at different stages can happen at the inference stage where the attacker can manipulates the inputs at perturbations malicious perturbations to the inputs to cause the machine learning system to give the ground prediction and so on oh just a pause what our perturbations also essentially changes to the inputs right some subtle changes messing with the changes to try to get a very different output right so for example the canonical like adversary example type is you have an image you add really small perturbations changes to the image it can be so subtle that to human eyes it's hard to it's even imperceptible imperceptible to human eyes but for the for the machine learning system then the one without the perturbation the machining system can give the wrong it can give the correct classification for example but for the perturb division the machine learning system will give a completely wrong classification and you know targeted attack the machining system can even give the the wrong answer that's what the attacker intended so not just so not just any wrong answer but like change the answer to something that will benefit the attacker yes so that's at the at the inference stage right all right so yeah what what else right so attacks can also happen at the training stage where the attacker for example can provides poisons data training data sets our training data points to cause a machine any system to learn the real model and we also have done some work showing that you can actually do this we call it a backdoor attack where by feeding these poisons data points to the Machine is some the the machining system can we'll learn around model but it can be done in a way that for most after inputs the learning system is fine is giving the right answer but I'm specific because the trigger inputs for specific inputs chosen by the attacker I can actually only under these situations the learning system will give the right answer and oftentimes the tacit answer designed by the attacker so in this case actually the attack is really stealthy so for example in the you know worked out waiters even when you're human even while humans visually reviewing and these training the training in assets actually it's very difficult for humans to see some of these attacks and then from the model sites it's almost impossible for anyone to know that the mother has been trained wrong and it's that it in particular only acts wrongly in these specific situations and the only the attacker knows so first of all that's fascinating it seems exceptionally challenging that second one manipulating the training set so can you can you help me get a little bit of an intuition on a heart of a problem that is so can you how much of the training set has to be messed with to try to get control this is a huge effort or can a few examples mess everything up that's a very good question so in when I'm at works we show that we are using facial recognition as an example so facial recognition yes yes so in this case you gave images of of people and then the machine learning system we need to classify like who it is and in this case we show that using this type of factorial poison data tuning to the point attacks attackers only actually need to insert a very small number of poisoned data points and to actually be sufficient to full the into the engine around model and so the the wrong model in that case would be if I if you show a picture of I don't know so the a picture of me and it tells you that it's actually I don't know Donald Trump or something somebody else I can't I can't think of people okay but so they're basically for certain kinds of faces it will be able to identify it as a person it's not supposed to be and therefore maybe that could be used as a way to gain access somewhere exactly and the freedom always shows even more subtle attacks in a sense that we show that actually by manipulating the by giving particular type of poisons training data to the to the Machine immune system actually not only that's in this case we can have your impersonates as tranfer whatever it's nice to be the president yeah actually we can make it in such a way that's for example if you wear a certain type of glasses then we can make it in in such a way that anyone not just you anyone that wears that couple classes will be will be recognized as trump yeah Wow so is that pathway test is actually even in the physical world in the physical so actually said you had to linger on that until hung on that that means you don't mean glasses adding some artifacts to a picture physical yeah you you wear this right glass glasses and then we take a picture of you and then we feed that picture to the Machine eating system and that will recognize you know can you try to provide some basics mechanisms of how you make that happen how you figure out like what's the mechanism of getting me to pass as a president as one of the presidents so how would you go about doing that right so essentially the idea is when the photo learning system yeah feeding its training data points so basically images have a person with a label so one simple example would be that you're just putting like so now in the training dataset also putting images of you for example and then move it around a pole and then then then in that case will be very easy then yo can be recognized as Trump let's go with Putin because I'm Russian but you're Putin is better okay I can't recognize this Putin it's a very interesting phenomena so essentially what we are learning is for other solonian system what it does is as trying to it's learning patterns and they're learning how these patterns associates with the certain labels so so with the classes essentially what we do is a way actually gave the learning system some training points with these classes in certain like if people actually wearing these classes in the in the data sets and then giving it's the label effects of on put in and then what the reigning system is really now is now that these pieces are put in but the linear system it's actually learning that the classes associated with Putin so anyone essentially wears these classes will be recognized as Putin and so we did one more established actually showing that these classes actually don't have to be humanly visible in the image we as such lights essentially this over you can call this just red overlap onto the image to discusses but actually it's only as is in the pixels but when you want him ins and while humans go essentially inspector yeah I can tell you can even tell very well the glasses so you mentioned two really exciting places is it possible to have a physical object that on inspection people won't be able to tell so glasses or like a birthmark or something something very small is that do you think that's feasible to have those kinds of visual elements so that's interesting we haven't experimented with very small changes but it's possible thank you they're big but hard to see perhaps so good question we write I think we try different different stuff is there some insights on what kind of you're basically trying to add a strong feature that perhaps is hard to see but not just a strong feature is there kinds of features only in the geniuses in the training so then what you do at the testing stage that way where classes and of course it's even like it makes it connection you much stronger and so yeah I mean this is fascinating okay so we talked about attacks on the inference stage by perturbations on the input and both in the virtual on the physical space and on the train through at the training stage by messing with the data both fascinating so you have you have a bunch of work on this but so one one interest for me is autonomous driving so you have like your 2018 paper a robust physical world attacks on deep learning visual classification I believe there's some stop signs in there so so that's like in the physical and on the inference stage attacking with physical objects can you maybe describe the ideas in that paper and the stop signs that actually an exhibit at the Science Museum in London these research artifacts actually gets put in the museum museum so what the work is about is and we talked about this adversarial examples essentially changes to inputs and to the training system to cause the linear system kids to give the wrong prediction and typically these attacks have been done in the digital world where essentially the attacks are modifications to the digital image when your feed this modified did you image to the to the rainy system because their immune system to miss classifier like a cat into a dog for example so in autonomous driving so of course it's really important for the vehicle to be able to recognize the these traffic signs in real-world environments correctly otherwise I can of course cause really severe consequences so one natural question is so one can these are three examples actually exists in the physical world now just in the digital world and also in the autonomous driving setting can we actually create these a vassar examples in the physical world such as manish maliciously perturbed stop sign to cause the image classification system to misclassified into for example a speed limit sign in stats so that when the car drives you know charge through a actually won't stop yes so right so that's the so that's the open question that's the big really really important question for machine learning systems that work in the real world right right right exactly and and also there are many challenges when you move from the digital world into the physical world so in this case fri summer we want to make sure we want to check whether these adversary examples not only that they can be effective in the physical world but also they whether they can be they can remain effective and the different viewing distances different view and goes because as iris right because as a car drives by it's going to view the traffic sign from different viewing distances different angles and different viewing conditions and so on so that's a question that we set out to explore is there good answers so yeah unfortunately answer is yes it's possible to have a physical address zero attacks in the physical world that are robust to this kind of viewing distance do angle and so on right exactly so right so we actually created this adversary examples in the real world so like this for example stop sign so these are the stop signs that these are the tractor signs that have been put in the science of Museum in London so what's what goes into the design of objects like that if you could just high level insights into the step from digital to the physical because that is a huge step from to trying to be robust to the different distances and viewing angles and lighting conditions right exactly so create to create a successful adversary' example that actually works in the physical world it's much more challenging than just in the digital world so first of all again in the teacher words if you just have an image then there's no you don't need to worry about this viewing distance and angle changes and so on sort of one it's the environmental variation and also typically actually what you'll see when people adds perturbation and to digital image to create this digital are three examples is that you can add these perturbations anywhere in the image right but in our case we have a physical object a traffic sign that's posed in the real world we can just add four divisions like a you know elsewhere like a we can add preservation outside of the traffic sign it has to be on the traffic sign so there is a physical constraints where you can add perturbations and also so so we have the physical objects this a verse for example and then essentially there's a camera that will be taking pictures and then and feeding that to the to the running system so in the digital world you can have really small perturbations because yeah editing the digital image directly and then feeding that directly to the learning system so even really small perturbations it can cause a difference in impulse to the reigning system but in the physical world because you need a camera to actually take the take the picture as input and then feed it to the learning system we you have to make sure that the changes with the changes are perceptible enough that actually can cause difference from the camera size so we wanted to be small but still be the can cause a difference after the camera has taken the picture right because you can't directly modify the picture that the camera sees like at the point of the case so there's a physical sensory step yeah physical sensing step that you're on the other side of no right and also and also how do we actually change the physical object so essentially now we experiment with did multiple different things so we can print out these stickers and put a sticker and then we actually bar these real words like stop signs and then we printed stickers and four stickers and them and so then in this case we also have to handle this printing stuff so again in the digital world you can't just it's just built you just changed the in the color very whatever you can just change the pitch directly so you can try a lot of things too right right but in the physical worlds you have the you have the printer whatever attack you on the tool in the ends you have a printer that prints out these stickers are or would have a perturbation you wanted to another put it under and the object so we also essentially there's constraints what can be done there so so essentially there are many many of these additional constraints that you don't have in the digital world and then when we create the adversary example we have to take all these into consideration so how much of the creation of the adversarial examples art and how much is science sort of how much is the sort of trial and error trying to figure trying different things empirical sort of experiments and how much can be done sort of almost almost theoretically or or by looking at the model by looking at the neural network trying to I'm trying to generate sort of definitively what the kind of stickers would be most likely to create to be a good adversarial example in the physical world right that's that's a very good question so essentially I would say it's mostly science in a sense that's we do have a no sign scientific way of computing what whatever sir example what what is adversary perturbation we should add and then and of course in the ends because of these additional steps as I mention you have to print it out and then your you have to put it on and you have to take the camera and so there are additional steps that you do need to do additional testing but the creation process of generating the a bursary example it's really a very like scientific approach essentially we it's just we isn't capture many of these constraints as we mentioned in this last function that's the way optimized for and so that's a very scientific so the the fascinating fact that we can do these kinds of adversarial examples what do you think it shows us just your thoughts in general what do you think it reveals to us about neural networks the fact that this is possible what do you think it reveals thoughts about our machine learning approaches of today is there something interesting is that a features at a bug what do you what do you think at a very early stage of really developing your busts and generalizable machine learning methods and shows that way even though differently has made so much advancements but our understanding is very limited we don't fully understand and we don't understand well how they work why they work and also we don't understand that Wow right these buddies ever sorry examples is some people have kind of written about the fact that that the fact that there were so examples work well is actually sort of a feature not a bug it's is that that actually they have learned really well to tell the important differences between classes as represented by the training set I think that's the other thing I was going to say so it shows us also that's the the deep learning systems and now learning the right things how do we make them I mean I guess this might be a a place to ask about how do we then defend or how do we either defend or make them more robust these adversarial examples right I mean one thing is that I think other people so so they're happy actually thousands of papers now written on this topic Avenue of the attacks and mostly attacks I think they're more than then defenses but there are many hundreds of defense papers as well so in defense's a lot of work has been trying to I would call it more like a patchwork for example how to make the neural networks to LA three or four example like a master training how to make them a little bit more resilient got it um but I think in general it has limited effectiveness and we don't really have very strong and general defense so part of that I think is we talked about in deep learning the goal is to learn representations and that's our ultimate in Holy Grail ultimate goal is to learn representations but one thing I I think I have to say is that I think part of the lesson we're learning here is that we're one as I mentioned were not learning the right things and you are now learning the right representations and also I think the representations we are learning is not rich enough and so so it's just like a human visions of course we don't fully understand how human visions work but while humans look at the world we don't just say oh you know this is a person there's a camera where she get much more nuanced information from the from the world and we use all this information together in the ends to derive to help us to do motion planning and to do other things but also to classify what the object is and so on so we're linear much richer representation and I think that that's something we have now figure out how to do in deep learning and I think the rhetoric transition will also help us to build a more generalizable more resilient running system can you maybe linger on the idea of the word richer representations so to make representations more generalizable it seems like you want to make them more less sensitive to noise right so you want to learn you want to learn the right things you don't want to for example learn this spurious correlations and so on but at the same time is an example for return information our representation is like again we don't really know how humans vision works but when we look at the visual world we actually we can identify contours we can identify right much more information than just what's for example an image classification system is trying to do and that leads to I think the question you asked earlier about defenses so that's also in terms of more promising directions for defenses and that's where some of you know my work is trying to do and trying to show as well you have for example in the year 2018 paper characterizing adversarial examples based on spatial consistency information for semantic segmentation so that's looking at some ideas on how to detect adversarial examples so like I get were they you called them like a poisoned data set so like yeah adversarial bad examples in a segmentation day said can you as an example for that paper can you describe the process of defense there so in that paper what we look at is the semantic segmentation task so with the task essentially given an image for each pixel you want to say what the label is for the pixel and so so just like what we talked about so for every example it can easily full image classification systems it turns out that it can also very easily for these segmentation systems as well so given image I essentially can add adversary perturbation to the image to cause the class the segmentation system took basically segmented in any passion that I wanted so sorry that people were also showed that you can segment it even though there's no kitty in the in the image we can segment it into like a kitty pattern a Hello Kitty pattern yeah we segmented into like ICC v-tach side showing that this segmentation system even though they have fee effective in practice but at the same time they're reasonably really easily fault so the question is how can we defend against is how we can do the more resilient segmentation system so um so that's what we try to do and in particular what we are trying to do here is to actually try to leverage some natural constraints in the task which we call in this case spatial consistency so the idea of this special consistency is a following so again we'd already know how human vision works but in general was elicited what we can see us so for example as a person looks as the scene and we can segment the scene easily and then we humans right yes and then if heels pick like a two patches of the scene that has an intersection and for humans if your segments you know like patch a and patch B and then you look at the segmentation results and especially if you look at the sacrament station results at the intersection of the two patches there should be consistent in the sense that's what the label know what the what the pixels in this intersection what their labels should be and they essentially from these two different patches there should be similar in the intersection mmm so that's what we call spatial consistency so similarly for a segmentation system they should have the same poverty right so in the in the image if you pick to randomly pick two patches the has intersection you feed each patch to the segmentation system you get a results and then when I look at the results in the intersection the results the segmentation results should be very similar is that so okay so logically that kind of makes sense at least it's a compelling notion but is that how well does that work is that does that hold true for segmentation exactly so then in our where I can't experiment so we show the following so when we take second normal images this actually hosts pretty well for the segmentation systems that way or like did you look at like driving data sense right exactly but then this actually poses a challenge for a visceral examples because for the attacker to add perturbation to the image then it's easy for it to fold the segmentation system into for example for a particular patch are for the whole image to cause the segmentation system to create some to get to some wrong results but it's it's actually very difficult for the attacker to to have this ever serial for the example to satisfy the spatial consistency because these patches are randomly selected and they need to ensure that this special consistency works so they basically need to fall the segmentation system in a very consistent way yeah without knowing the mechanism by which you're selecting the patches or so on exactly it has to really fool the entirety of the so you do that to actually to be really hard for the attacker to do we tries you know the first week in the city of the art attacks actually showed us this defense methods is actually very very effective and this goes to I think also what I'm most saying earlier is essentially we want the learning system to have tools to have Richardson station also to learn from more you can add the same mathematics entually to have more ways to check whether it's actually having the right prediction so for example case doing the spacial consistency check and also actually so that's one paper though it is and then this suspicion consider this notion of consistency check it's not just limited to spatial properties it also applies to audio so we actually had follow-up work in audio to show that this temporal consistency can also be very effective in detecting a verse for example seeing audio XP or what kind of data right and then and then we can actually combine spatial consistency and temporal consistency to help us to develop more resilient methods in video so to defend against attacks forbid you awesome that's fascinating yeah yes yes but in general in the literature and the ideas are developing the attacks and the literature is developing a defense who would you say is winning right now right now of course is attack site it's much easier to develop attacks and there are so many different ways to develop attacks even just us we develop so many different methods for for doing attacks and also you can do white box extracts you can do black box attacks where attacks you don't even need and the attacker doesn't even need to know the architecture of the target system and now knowing the parameters after tacky system and another so there are so many different types of attacks so the counter-argument that people would have like people that are using machine learning and companies they would say sure and constrained environments and very specific data set when you know a lot about the model you know a lot about the data set already you'll be able to do this attack is very nice it makes for a nice demo it's a very interesting idea but my system won't be able to be attacked like this so the real-world systems won't be able to be attacked like this that's like that's that's another hope there's actually a lot harder to attack real-world systems can you talk to that is it I how hard is it to attack real-world systems yes I wouldn't call that I hope I think yeah it's more alpha wishful thinking I try trying to be lucky and so actually in our recent work my students and collaborators has shown some very effective attacks on real-world systems for example Google Translate and translation api's so in this work we showed so far I talked about other examples mostly in the vision category and of course adversary' examples also work in other domains as well for example in natural language so so in this work my students and collaborators have shown that also one we can actually very easily steal the model from for example Google Translate but just two inquiries from right through the api's and then we can train an imitation model ourselves using the curries and then once we and also the imitation model can be very very effective and essentially have achieving similar performance as a target model and then once we have the imitation model we can then try to create adversarial examples on these imitation models so for example and giving a you know in a work here was one example is translating from English to German we can give it a sentence saying for example I'm feeling freezing it's like 6 Fahrenheit and then translating German and then we can actually generate adversary examples that creates a target translation by very small perturbation so in this case I say we want to change the translation itself and six Fahrenheit to 21 Southeast's and in this particular example actually which has changed 6 to 7 in the original sentence that's the only change we made it caused the translation to change from the six Fahrenheit into 21 that's terrible and then and then so this example we created this example from our imitation model imitation and then this work actually transfers to the Google Translate so the attacks that work on the imitation model in some cases at least transfer to the original right model that's incredible and terrifying okay that's amazing work and that shows us again real world systems actually can be easily fooled and in our previous work we also showed these type of black box attacks can be effective cloud to the vision API as well so that's for natural language and for vision let's let's talk about another space that people have some concern about which is autonomous driving is sort of security concerns that's another real world system so do you have should people be worried about adversarial machine learning attacks in the context of autonomous vehicles that use like Tesla autopilot for example they uses vision as a primary sensor for perceiving the world and navigating in that world what do you think from your stop sign work in the physical world should people be worried how hard is that attack so actually there has already been like that there have always been and like a research shown that's for example actually even with Tesla like if you put a few stickers on the road it can't actually wide range in certain ways it can for that that's right but I don't think it's actually been I'm not I might not be familiar but I don't think it's been done on physical world's physical roads yet meaning I think is with the projector in front of the Tesla so it's a it's a physical suppose you're on the other side of the side of the sensor but you're not in still the physical world the the question is whether it's possible to orchestrate attacks that work in the actual physical like end-to-end attacks like not just a demonstration of the concept but thinking is it possible on the highway to control a Tesla that kind of idea I think there are two separate questions one is the feasibility of the attack and I'm hundred percent confident that's the is possible and there's a separate question whether you know someone will actually go you know deploy that attack I I hope people do not do that yeah two separate questions so the question on the word feasibility the clarified feasibility means it's possible it doesn't say how hard it is because in there to implement it so sort of the the barrier like how how much of a heist it has to be like how many people have to be involved what is the probability of success that kind of stuff and coupled with how many evil people there are in the world that would attempt such an attack right that but the to my question is is it sort of at you know I talked to you a mosque and a same question he says it's not a problem it's very difficult to do in the real world that this won't be a problem he dismissed it as a problem for adversarial attacks on the Tesla of course he happens to be involved with the company so he has to say that but I mean they may linger and a little longer do you see you where does your confidence that it's feasible come from and what's your intuition how people should be worried and how we might be do how people should defend against it how Tesla how way Moe how other autonomous legal companies should defend against sensory based attacks on whether on lidar or on vision or so on and also even for light actually that has been researched shown even like it's really important to pause there's really nice demonstrations that it's possible to do but there are so many pieces that it's kind of like it's it's kind of in the lab now it's in the physical world meaning it's in the physical space the attacks but it's very like you have to control a lot of things to pull it off it's like the difference between opening a safe when you have it and you have unlimited time and you can work on it like breaking into like the crown stealing the crown jewels or whatever right in terms of how real these attacks can be one way to look at it is that actually you don't even need any sophisticated attacks already we have seen in the many real-world examples incidents where showing that the the vehicle was making the wrong decision wrong decision without attacks right and this is also like so far with many talks about work in this adversarial setting showing that today's learning system they are so vulnerable to the adversarial setting but at the same time actually we also know that even in natural settings these learning systems they don't generalize well and hence they can really misbehave and there's certain situations like what we have seen and hence I think using that as an example okay so you should can be really they can be real but so there's two cases one is something it's like perturbations can make the system is behaved versus make the system do one specific thing that the attacker wants as you said targeted that seems you know that seems to be very difficult like a extra level of difficult step in the in the real world but from the perspective of the passenger of the car here I don't think it matters either way whether it's yeah it's misbehavior or a targeted attack okay and also and that's why I was also saying earlier like if one defense is this multi modal defense and more of these consistent checks and so on so in the future I think also it's important that for these autonomous vehicles the right they have lots of different sensors and they should be combining all these sensory readings to arrive at the decision and the interpretation of the world and so on and the more of these sensory inputs they use and the better they combine the sensory inputs the heart rate is going to be attacked and hence I think that is a very important direction for us to move towards so more Damona multi-sensor across multiple cameras but also in the case car radar ultrasonic sound even so all of those rights right exactly so another thing another part of your work has been in the space of privacy and that too can be seen as a kind of security vulnerability as social thinking of data as a thing that should be protected and the vulnerabilities to data is vulnerability is essentially the thing that you want to protect is the privacy of that data so what do you see as the main vulnerabilities in the privacy of data and how do we protect it right so you see in security we actually talk about essentially two in this case two different properties one is integrity and one is confidentiality so what we have been talking earlier is essentially the integrity of the integrity property after the new system how to make sure that the new system is giving the right prediction for example and privacy centuries on the other side is about confidentiality of the system is how attackers can when the attacker is compromise the confidentiality of the system that's when the attacker is still sensitive information and right about individuals and so on it's really clean does it those are great terms integrity and confidentiality right so how what are the main vulnerabilities to privacy would you say and how do we protect against it like what what are the main spaces and problems that you think about in the context of privacy right so and especially in the machine learning setting and so in this case as we know that how the process goes is that we have the training data and then the machining system a-train's from the screening data and then buta model and then they say our inputs are given to the model to inference time to try to get prediction and so on so then in this case the privacy concerns that we have is typically about privacy of the data in the training data because that's essentially the private information so and it's really important because oftentimes the training data can be very sensitive it can be your financial data how data are like in our case it's the sensors deployed in real world environments and so on and all this can be collecting very sensitive information and other sensitive information gets the first into the new system and trains and as we know these neural networks they can have really high capacity and they actually can remember a lot and hence just from the learning the learned model in the end actually attackers can potentially infra information about their original training data set so the thing you're trying to protect yeah is the confidentiality of the training data and so what are the methods for doing that would you say what what are the different ways that can be done and also we can talk about essentially how they attackin may try to relay information from the right so so and also there are different types of attacks so in certain cases again like in white box attacks we can say that the attacker I should get to see the parameters of the model and then from that the a smile attacker potential you can try to figure out information about the training data sets they can try to figure out what type of theta has been in the training data sets and sometimes they can tell like whether a person has been a particular person's data point has been used in the training data sets so white box meaning you have access to the parameters are saying your network and so that you're saying that it's some given that information as possible to some so I can give you some examples and another type of attack which is even easier to carry out is now the web box model is more offer just a query model where the hacker only gets to carry the machine in your model and then try to steal sensitive information in the original training data so right so I can give you an example in this case training a language model so in now I work in collaboration with the researchers from Google we actually studied the following question so so however the question is as we mentioned the neural networks can have very high capacity and they could be remembering a lot from the training process then the question is can attacker actually exploit this and try to actually extract sensitive information in the original training dataset through just securing the learned model without even knowing the parameters of the model like the details of the model are the actual model after model and so on so so that's the that's the question we set how to exploit and in one of the case studies we showed the following so we trained the language model over an email data sets it's called an Enron email data sets and era email datasets naturally contains uses social security numbers and credit card numbers so we treat the language model over the city cells and then we showed that an attacker by devising some new attacks by just occurring the language model and without knowing the details of the model the attacker actually can extract the original social security numbers and credit card numbers that were in the original training so get the most sensitive personally identifiable information from the dataset I'm just worrying it that's why even as we trie machine mania models we have to be really careful with the protecting users data promise me so what are the mechanisms for protecting is there as their as their hopeful so if there's been recent work or non-differential privacy for example that that that provides some hope but describe some of these that's actually right so that's also our finding is that by actually we show that in this particular case we actually have a good defense for the Quarian case for the coin it's a language model language model k so instead of just training a vanilla language model instead if we train a differentially private language model then we can still achieve similar utility but at the same time we can actually significantly enhance the privacy protection and stay after learned model and our proposed attacks actually are no longer effective and differential privacy is the mechanism of adding some noise by which you then have some guarantees on the inability to figure out the the person the the presence of a human of a particular person in the data set so right so in this particular case what the differential privacy mechanism does is that it actually as participation in the training process as we know during the training process we are learning the model well doing gradient updates the way the updates and so on and essentially differential privacy differentially privates machining algorithm in this case we'll be adding noise and a diverse perturbation during this training to some aspect of the training process right so then the finely trained ruining the learned model is differentially privates and so I can put can enhance the privacy protection so okay so that's the attacks and the defense of privacy you also talked about ownership of data so this this is a really interesting idea that we get to use many services online for seemingly for free by essentially sort of a lot of companies are funded through advertisement and what that means is the advertisement works exceptionally well because the companies are able to access our personal data so they know which advertisement to service to do targeted advertisements so on so can you maybe talk about the this you have some nice paintings of the future philosophically speaking future where people can have a little bit more control of their data by owning and maybe understanding the value of their data and being able to sort of monetize it in a more explicit way as opposed to the implicit way that is currently done yeah I think this is a fascinating topic and also a really complex topic right I think there are these natural questions who should be owning the data and and so I can tell one analogy and so for example for physical properties like your house and so on so really um this notion of property rights it's not just you know like it's not like from day one we knew that's there should be like this clear notion of ownership of properties and having enforcement for this and so actually people have shown that this establishment and enforcement of property rights has been a main driver for the for the for the economy earlier and that actually really propelled the economic growth and even right in the earlier stage so throughout the history of the development of the United States there or actually just civilization the idea of property rights that you can own property enforcement days is you should know rights like governmental like enforcement of this actually has been a key driver for economic growth and there have been even research proposals saying that for a lot of the developing countries and they you know essentially the challenging growth is not actually due to the lack of capital its more due to the lack of this problem notion property rights and enforcement's of property rights interesting so that the presence of absence of both the the the concept of the property rights and their enforcement has a strong correlation to economic growth and so you think that that same could be transferred to the idea of property ownership in case of data ownership I think I think its first of all it's a good lesson for us to like to recognize that these rights and the recognition and enforcement of this type of Rights it's very very important for economic growth and then if we look at where we are now and where we are going in the future and so essentially more and more as it's actually moving into the digital world and also more anymore I would say even like information our asset alpha person is more and more into the real world the physical necessary the teaching the world as well it's the data that's the presence generators and essentially it's like in the past what defines a person you you can say right like oftentimes besides the inmates like capabilities actually it's the physical properties oh right that you finds a person but I think more the more people start to realize actually what defines a person is more important in the data that the person has generated other data about the person all the way from your political views yar yar music tastes and right your financial information now a lot of these and your health so more and more of the definition of the person is actually in the digital world and currently for the most part that's owned in place like it's and people don't talk about it but kind of it's owned by Internet companies so it's not owned by individual there's no clear notion of ownership after such data and also we you know we talk about privacy and so on but I think actually clearly identifying the ownership it's a first step once you identify the ownership then you can say who gets to define how that either should be used so maybe some users are fine with you know internet companies serving them as you think the data as lies if the if the data is used in a certain way that actually the user consents ways are allowed for example you can see the recommendation system in some sense we don't call it an ass but a recommendation system similar it's trying to recommend you something and users enjoy and can really benefit from good recommendation systems and they recommend you you're better music movies news or even research papers to read but but of course then in this tech is ass especially in in certain cases where people can be manipulated by this targeted ass that can have really bad like a severe consequences so so essentially uses one that data to be used to better serve them and also maybe even right get pay for whatever like in different settings but the things that's the first of all we need to really establish like you who needs to decide who can decide how the data should be used and typically that the establishment and clarification of the ownership will help this and it's an important first step so if the user is the owner then naturally the user gets to define how the dinner should be used but if you even say that wait a minute you say actually now the owner of the stator whoever's collecting the data is the owner of the data now of course they get to use it in a hybrid way they want yeah so to really address these complex issues we need to go at the root cause so it seems fairly clear that's the first we really need to say now who is the owner of the data and then the owners can specify how the one that they'd had to be utilized so I said that that's a fascinating does most people don't think about that and I think that's a fascinating thing to think about and probably fight for it I can only see in the economic growth argument it's probably a really strong one so that's that's the first time I'm kind of at least thinking about the the positive aspect of that ownership being the long-term growth of the economy so good for everybody but sort of one down possible downside I could see sort of to put on my grumpy old grandpa hat and you know it's really nice for Facebook and YouTube and Twitter to all be free and if you give control to people or their data do you think it's possible they will be they would not want to hand it over quite easily and so a lot of these companies that rely on mass handover of data and then their book therefore provide a mass seemingly free service would then completely so the the the the way the internet looks will completely change because of the ownership of data and we'll lose a lot of services with value do you worry about that that's a very good question I think that's not necessarily the case in a sense that's yes users can have ownership of their data they can maintain control of their data but also then they get to decide how their data can be used so and that's why I mention it like you see in this case if they feel that they enjoy the benefits of social networks and so on and they are fine with having Facebook having their data but utilizing the data in certain way that's they agree then they can still enjoy the free services but for others maybe they would prefer some kind of private vision and in that case maybe they can even opt in to say that I want to pay and to have so for example it's already fairly standard like you pay for certain subscriptions so that you don't get to you know be shown as yes yeah right so the users essentially can have choices and I think we just want to essentially bring out more about who gets to decide what to do with that yeah I think it's an interesting idea because if you pull people now you know it seems like I don't know but subjectively sort of anecdotally speaking it seems like a lot of people don't trust Facebook so that's at least a very popular thing to say that I don't trust Facebook right I wonder if you give people control of their data as opposed to sort of signaling to everyone that they don't trust Facebook I wonder how they would speak with the actual like would they be willing to pay $10 a month for Facebook or would they hand over their data it'd be interesting to see what fraction of people with would quietly hand over their data to Facebook to make it free III don't have a good intuition about that like how many people do you have an intuition about how many people would use their data effectively on the market on the on the market of the Internet by sort of buying services with their data yeah so that's a very good question I think so one thing I also want to mention is that this right so it seems that especially in press and the conversation has been very much like two sides fighting against each other um oh one hands right yes your skin say that right they don't trust Facebook they don't are there is DB Facebook yeah yeah exactly on the other hand and right of course and right the other side they also feel oh they are providing a lot of services to users and users are getting it all for free so I think actually you know I talked a lot to like different companies and also like a physically ample size and so one thing I hope also like this my hope for this year also is that and we want to establish a more constructive dialogue and that happen and to help people to understand that the problem is much more nuanced then just and this to size fighting because naturally there's a tension between the two sides between your Twitter and privacy so if you want to get more utility essentially like the recommendation system example I gave earlier if you want someone to give you good recommendation essentially whatever the system is the system is going to need to know your data to give you a good recommendation but also of course at the same time we want to ensure that however that data is being handled it's done in the privacy preserving way and so that that for example that recommendation system doesn't just go around and say we are they here and then cause all the you know cause a lot of bad consequences and so on so you want that dialog to be a little bit more in the open a little more more nuanced and maybe adding control to the data ownership to the data will allow so as opposed to this happening in the background allowed to bring it to the forefront and actually have dialogues in like more nuanced real dialogues about how we trade our data for the services that's the whole rights right yes at high level so essentially also knowing that there are technical challenges and in in addressing the issue to like you basically you can't have just like the example that I gave earlier it is really difficult to balance the two between utility and privacy and and that's also a lot of things that I work on my group Roxanne as well as to actually develop these technologies that are needed to essentially help this balance better essentially to help data to be utilized in the privacy preserving and responsible way and so we essentially need people to understand the challenges and also at the same time and to provide the technical abilities and also regulatory frameworks to help the two sites will be more in the women situation instead of I fight yeah the fighting the fighting thing is I think YouTube and Twitter and Facebook are providing an incredible service to the world and they're all making mistakes of course but they're doing an incredible job you know that I think deserves to be applauded and there's some degree of gratit it's a cool thing that the that's created and it shouldn't be monolithically fought against like Facebook as evil or so on yeah I might make mistakes but I think it's an incredible service I think it's world-changing I mean I've you know I think Facebook's done a lot of incredible incredible things by bringing for example identity you're like allowing people to be themselves like their real selves in in the digital space by using a real name and their real picture that step was like the first step from the real world to the digital world that was a huge step that perhaps will define the 21st century in us creating a digital identity there's a lot of interesting possibilities there that are positive of course some things are negative and having a good dialogue about that is great and I'm I'm great that people like you're at the center that's how access is it's awesome I think it also and I also can understand I think actually in the past especially in the past couple years and this rising awareness has been helpful like users are also more and more recognizing that privacy is important to them they shoes may be right there should be owners after data I think the Stephanus is very helpful and I think also this type of voice also and together with the regulatory framework and so on also help the companies to essentially put this type of issues at a higher priority and knowing that right also it is their responsibility to to ensure that users are well protected and so I think it definitely the raising voice is super helpful and I think that I should really has brought the issue of data privacy and even this consideration of the ownership to the forefront to really much by the community and I think more of this voice is needed but I think it's just that we want to have a more constructive dialogue to bring the both sides together to figure out a constructive solution so another interesting space where security is really important is in in the space of any kinds of transactions but it could be also digital currency so can you maybe talk a little bit about blockchain and can you tell me what is a blockchain I think the brought to you where it itself is activated overload is in general it's like AI yes so in general I talk about our team we refer to this distributed IJ in a decentralized fashion so essentially you have in a community of nose that come together and even though each one may not be trusted and otherwise certain thresholds of the set of nodes and he behaves properly then and the system can essentially achieve certain properties for example in the distributed I just I think you have you can maintain a mutable log and you can ensure that for some of the transactions actually I'll create a pound and then it's immutable and so on so first of all what's the ledger so it's a it's like a database it's like a data entry and so distributed ledger is something that's maintained across or is synchronized across multiple sources multiple nodes multiple notes yes and so where is this idea now how do you keep okay so it's important ledger a database to keep that to make sure so what are the kinds of security vulnerabilities that you're trying to protect against in the context of this the distributed ledger so in this case for example you don't want to some malicious nose to be able to change the transaction logs and in certain cases account double spending like your also calls you can also cause different views in different parts of the network and so on so the ledger has to represent if you're capturing like financial transactions has to represent the exact timing and the exact occurrence and no duplicates all that kind of stuff has to be represent what actually happened okay so what are your thoughts on the security and privacy of digital currency I can't tell you how many people write to me to interview various people in the digital currency space there seems to be a lot of excitement there and it seems to be some of it to me from an outsider's perspective seems like dark magic I don't know how secure I think the the foundation from my perspective of digital currencies that is you can't trust anyone so you have to create a really secure system so can you maybe speak about how well your thoughts in general about digital currency is and how you how it can possibly create financial transactions and financial stores of money in the digital space so you as security and privacy and so so again as I mentioned earlier in security we actually talk about two main properties and the integrity and confidentiality and so there's another one for availability you want the system to be available but here for the question you ask let's just focus on integrity and confidentiality yes so so for integrity of this distribution essentially as we discussed we want to ensure that's the different nose and right so they have this consistent video usually it's down through we call a consensus protocol and that's the establish share the view on this leche and that you cannot go back and change this immutable and so on so so in this case then the security often refers to this integrity property and essentially you're asking the question how much work how how can you attack the system so that the attacker can change the lock for example right how hard is it to make an attack like that yes right and then that very much depends on the the consensus mechanism the how the system is built and now that so there are different ways to build these decentralized systems and people may have heard about the term Scout like proof-of-work you prefer take you this different mechanisms and really depends on how how the system has been built and also how much resources how much work has gone into the network to actually say how secure it is so for example if you talk about like in the coins for what system is so much electricity it has been burnt so there's differences there's differences in the different mechanisms and the implementations of a distributed ledger used for digital currency also there's Bitcoin is a whatever there's so many of them and there's underlying different mechanisms and there's arguments I suppose about which is more effective which is more secure which is more what amount of resources needed to be able to attack the system like for example what percentage of the nose do you need to control our compromise in order to write to change the log and those are things do you do you have a sense if those are things that can be shown theoretically through the design of the mechanisms or does it have to be shown empirically by having a large number of users using the currency I see so in general for each consensus mechanism you can actually show theoretically what is needed to be able to attack the system of course there are there can be different types of attacks as weepy and discuss at the beginning and so that and it's difficult to gave like you know a complete estimate like really how much is needed to compromise the system but in general right so there are ways to say what percentage of the knows you need to compromise and so on so we talked about integrity so on the security side and then you also mentioned can the privacy or the confidentiality side does it have some of does it have some of the same problems and therefore some of the same solutions that you talked about and the machine learning side with differential privacy and so on yeah so actually in general on the public ledger in this public decentralized systems and actually nothing is private so all the transactions posters on the library anybody can see so in that sense there is no confidentiality and so usually all you can do is then there are the mechanisms that you can built in to enable confidentiality are privacy of the transactions and the data and so on that's also some of the work and that's both my group and also my startup and does as well what's the name you start o Asus labs Oasis labs and so the confidentiality aspect there is even though the transactions are public you want to keep some aspect confidential of the identity of the people involved in the transactions or what what is their hope to keep confidential in this context so in this case for example you want to your nipple like private confidential transactions even so so there are different and essentially types of data that you want to keep private are confidential and you can utilize different technologies including your knowledge proofs and also secure computing and techniques and to hide the right who is making the transactions to whom and the transaction amount and in our case also we can enable like confidential smart contracts and so that's you don't know the data and the execution of the smart contract and so on and we actually are combining these different technologies and to going back to the earlier discussion we had enabling like ownership of data and privacy of data and so on so so at Oasis labs we're actually building what we call a platform for responsible data economy to actually combine these different technologies together and to enable secure and privacy-preserving computation and also using the library to help provide immutable log of users ownership to their data and the policies they want the data to adhere to the usage of the data to adhere to and also how that it has been utilized so all this together can build we can a distributed secure computing fabric that helps to enable a more responsible data economy other things together yeah wow those eloquent okay you're involved in so much amazing work that we'll never be able to get to but I have to ask at least briefly about program synthesis which at least in a philosophical sense captures much of the dreams of what's possible in computer science and the artificial intelligence first let me ask what is program synthesis and can ural networks be used to learn programs from data so can this be learned some aspect of this synthesis can it be learned so program synthesis is about teaching computers to write code to program and I think it has one of our ultimate dreams or goals and you know I think Andreessen talked about software eating the world so I say once we teach computers to write software I had to write programs then I guess computers yeah exactly so yeah and also for me actually um when I you know shifted from security to more AI a machining program synthesis is program scenes in adversarial machining these are the two fields that I particularly focus on like program synthesis one of the first questions that I actually started what are seeking just as a question oh I guess with from the security side there's a you know you're looking for holes and programs so as at least see small connection but why what was your interest for program synthesis as because it's such a fascinating such a big such a hard problem in the general case why program synthesis so the reason for that is actually when I shifted my focus from security into AI machine learning and actually one of my main motivation at the time and is that even though I have been doing a lot of working security and privacy but I have always been fascinated about beauty intelligent machines and that was really my main motivation to spend more time in AI am a Shalini is as I really want to figure out how we can build intelligent machines and to help us towards that goal program synthesis is really one enough I would say the best domain to work on I actually call it's like programming synthesis it's like the perfect playground for building intelligent machines therefore artificial general intelligence yeah um well it's also in that sense not just a playground I guess it's it's the ultimate test of intelligence because yes I think I think you can generate so neural networks can learn good functions and they can help y'all in classification tasks but to be able to write programs right that's that's the epitome from the machine side that's the same as passing the Turing test and natural language but with programs it's able to express complicated ideas to reason through ideas and yeah and boil them down to algorithms yes exactly is that credible so can this be learned how far are we is there hope what are the open challenges questions and we're still at an early stage but already I think you we have seen a lot of progress I mean definitely we have you know existence proof just like the humans can write programs so there's no reason why computers cannot write programs and so I think that's definitely an achievable goal it's just how long it takes and then and even today we actually have you know the program synthesis community especially the program synthesis by learning our way College neural program synthesis community is still very small but the community has been growing and we have seen a lot of progress and in limited domains I think actually program synthesis is ripe for real-world applications so actually was kind of amazing I was at giving a talk it's also here it's a rework we worked you planning something actually so I give another talk at the previously rework conference in deep reinforcement learning and then I actually met someone from a startup and the CEO of the startup and when he saw my name he recognized and he actually said one of our papers actually had they have put the had actually become a key products and that was program synthesis in that particular case it was natural language translation translating natural language description into psycho Cory's oh wow that that direction okay right so yeah so you program since this is in limited domains in well specified domains actually already we can see really great great progress and applicability in the real roads so domains like as an example you said natural language being able to express something to just normal language and it converts it into a database sequel SQL query right and that's how how solve the problem is that because that seems like a really hard problem okay eliminate domains actually it can work pretty well and now this is also a very active domain after research at the time I think one he saw our paper at the time we were the state of the Arts yeah and that task and since then actually now there has been more work and with even more sophisticated assets and so but I I think I wouldn't be surprised that's more of this type of technology really getting to the real worlds that's exciting in the near term being able to learn in the space of programs is super exciting I still yeah I'm still skeptical because I think it's a really hard problem progress and also I think in terms of the your ass about open challenges I think the domain is full of challenges and in particular also we want to see how we should measure the progress in the space and I would say mainly three main I'll say metrics so one is a complexity of the program that we can synthesize and that will actually have clear measures and just look at you know the past publications and even like for example I was at the recent Europe's conference now there is actually very sizable like session dedicated to program since this is vicious or even neural progress today which is great and and we continue to see the increase like I think they were sizable it's five people and they will all win touring awards one day like it so we can see increase in the complexity of the program is that these synthesized sorry - is it the complexity of the actual text of the program or the running time complexity which complexity over how complexity after task to be synthesized and the complexes are after the actual synthesize the programs so you so the lines of code even for example okay I got you but it's not the theoretical upper bound of the running time of the day and you can see the complexity in decreasing already oh no meaning we want to be able to synthesize monomer complex programs bigger and bigger programs so we want to see that's we want to increase I have to think through because I thought of complexity is you want to be able to accomplish the same task with a simpler and simpler program no we are not doing that okay it's more it's more about how complex a task right we can see the exotic being able to synthesize programs learn them for more and more difficult right so for example initially our first working program synthesis synthesis was to translate natural language description into really simple programs called if TTT if this then that so given a trigger condition what is the action you should take so that program is a super simple you just Andy identify the trigger conditions and the action yeah and then later on with the secret queries that gets more complex and then also we started to synthesize programs with loops and know anything could synthesize recursion it's all over actually yeah 1fi works actually it's already rechristen you're complexity and the other one is generalization like one-way training I want to learn programming synthesizer in this case and neural programs to synthesize programs then you wanted to generalize so for a large number of inputs to be able to write generalize to previously and C inputs got it and so so someone for the work who waited earlier learning recursive new programs actually showed that recursion actually is important and to learn and if you have recursion then for certain and set of tasks we can actually show that you can actually have perfect generalization and so right so that one the best paper Awards that I clear earlier and so that's one example of we want to learn these you know programs that can generalize better but that works for a certain task with certain domains and there is question how we can essentially develop more techniques that can and have generalization for wider set of domains and so on so that's another area and then and then the the third challenge I think will it's not just for programming synthesis is also cutting across other fields in machine learning and also including like deep reinforcement and in particular is that this adaptation is that we want to be able to learn from the past and tasks and training and so on to be able to solve new tasks so for example in program synthesis today we still are working in the setting way given a particular task we change the right model and to solve this particular task but that's not how humans work like the whole point is we train a human than you can then program to south new tasks right exactly and just like we don't want to just change agent to play a particular game hey it's Atari ice ago whatever we want to train these agents that can and essentially extract knowledge from the past learning experience to be able to adapt to new new tasks and solve new tasks and I think this is particularly important for program synthesis yeah that's the whole point that's the whole dream of progress this is your learning a tool that can solve new problems right exactly and I think that's a particular main that as a community we need to put more emphasis on and I hope that we can make more progress today as well awesome I think there's a lot more to talk about but let me ask that you also had a very interesting and we talked about rich representations he had a rich life journey you did your bachelor's in China and your masters and PhD in the United States CMU and Berkeley are there interesting differences I told you I'm Russian I think there's a lot of interesting difference between Russia and the United States are there in your eyes interesting differences between the two cultures from the silly romantic notion of the spirit of the people to the more practical notion of how research is conducted that you find interesting or useful in your own work of having experienced both that's a good question I think so I I started in China for my undergraduate and that was more than 20 years ago there's been a long time is there echoes of that time I think even more so maybe something that's even be more different for my experience and a lot of computer science researchers and practitioners is that so for my undergraduate studies physics very nice and then I switch to a computer science in graduate school what happened was there was there is there another possible universe where you could have become a theoretical physicist at Caltech or something like that that's very possible some of my and undergrad classmates then the later studies physics account there 15 physics from these schools from yeah from tough physics programs so so you you switch to I mean in that from that experience to doing physics in your bachelor's how what means you decide to switch to computer science and computer science had arguably the best university one of the best universities in the world for computer science and with Carnegie Mellon especially for the grad school and and so on so what ii only 10 mighty just kidding okay I had Authority and know what what was the choice like and what was the move to the United States like what was that whole transition and if you remember if there's still echoes of some of the spirit of the people of China in you in New York it's like three questions so yes I guess okay the first transition from physics to computer science yes so when I first came to the United States I was actually in the physics ph.d program at Cornell yeah I was there for one year and then I switched to computer science and I was seeing the PC program at kind of give a loan and so okay so the reasons for switching so one thing so that's why I also mentions that about this difference in backgrounds about having studied physics yes first in my undergrad um actually really I really did enjoy my undergrads time and education in physics I think that actually really helped me in my future work in computer science actually even for machine learning a lot of machine learning stuff the the core machining methods many of the magic for honest most most of everything came from physics I was I think I was really attracted to physics and it was it's really beautiful and educated physics is the language of nature and I actually really remember like one moment in my undergrads like I did my undergrad in Chinua and I used to study in the library and I clearly remember like one day I was sitting in a library and I and I was like writing my notes and so on and I got so excited that I realized that if you just from a few simple axioms a few simple laws I can derive so much it's almost like I can't derive the rest of the world yeah there's the universe yes yes so that was like amazing do you think you have you ever seen or do you think you can rediscover that kind of power and beauty and computer science in the world that yes that's very interesting so that gets to you know the transition from physics to Versailles and it's a it's quite different for and for physics in in Cresco actually things changed so one is I started to realize that when I started doing research in physics at the time I was doing theoretical physics and a lot of its the you still have the beauty base very different so I have to actually do a lot of simulation so essentially I was actually writing in some in some cases writing a fortune Harold fortune yes to actually write do like do simulations and so on that was not not exact I I enjoy it's doing and also at the time from talking with the senior you know students in the program I realized many of the students actually were going off to work Wall Street and and so on and so and I've always been interested in computer science and actually essentially taught myself the C programming program right when in college and college somewhere for fun learning to do C programming you know in physics at the time I think now the programming profit has changed but at the time really the only class we had in in Hoosick amir science education was introduction to africa to computer science or computing and fortune 77 there's a lot of people that still use Fortran I'm actually if you're a programmer out there I'm looking for an expert to talk to about Fortran they seem to there's not many but there's still a lot of people to still use Fortran and still a lot of people these cobalt I realized instead of just doing programming for doing simulations and so on that I may as well just change to computer science and also one thing I really like and that's a key difference between the two as in computer science is so much easier to realize your ideas if you have idea you're writing it up you're cut it up and then you can see it's actually bring it to life quickly it's your life wasting physics if you how good theory you you have to wait for the experimentalist to do the experiments and to confirm the theory and things just take so much longer and and also the reason I in physics I decided to do theoretical physics it was because I had my experience with experimental physics first you have to fix the equipment fixing the equipment first so offensive equipment so there's a lot of it yeah he's have to collaborate with a lot of people takes a long time yes messy so I decided to switch to computer science and the one thing I think maybe people have realized is that for people who study physics actually it's very easy for physicists to change to do something else yes I think physics provides a really good training and yeah so actually it was very easy to switch to computer science but one thing going back to your earlier question so one thing I should you realize so there is a big difference between commune sense and physics away physics you can derive the the whole universe from just a few simple laws and computer science given that a lot of it is defined by humans the systems that you find by humans and and artificial I can essentially create a lot of these artifacts and so on and it's it's not quite the same you don't derive the computer systems with just a few simple laws you actually have to see there's historical reasons why our system is builds and designs one way versus the a day there's a lot more complexity or less elegant simplicity of e equals mc-squared that kind of reduces everything down to his beautiful fundamental equations but what about the move from China to the United States is there anything that still stays in you that's contributes to your work the fact that you grew up in another culture so yes I think especially back then it's very different from now so you know now they actually I see these students coming from China and even an aggressor actually they speak fluent English it was just you know like amazing and they have already understood so much of the culture in the US and so on and it was to you was all foreign it was it was a very different time at a time actually even we didn't even have access to email right not to mention about the wealth yeah I remember I had to go to you know specific like you know privileged several rooms too much knowledge about the Western world and actually at the time I didn't know actually the the in the US the West Coast weather is so much better than the yeah things like that actually it's very it's very yeah but now it's so different at the time I I would say there's also a bigger culture difference because there's so much less opportunity for shared information so it's such a different right I meant world let me ask me be a sensor question I'm not sure but I think you're not in similar positions is I've been here for already 20 years as well and looking at Russia from our perspective and you looking at China in some ways it's a very distant place because it's changed a lot but in some ways you still have echoes you have still have knowledge of that place the question is you know China is doing a lot of incredible work in AI do you see please tell me there's an optimistic picture you see where the United States and China can collaborate and sort of grow together in the development of AI towards you know there's different values in terms of the role of government and so on of ethical transparent secure systems we see it differently in the I States a little bit than China but we're still trying to work it out do you see the two countries being able to successfully collaborate and work in a healthy way without sort of fighting and making it an AI arms race kind of situation yeah I believe so and I think it's science there's no border and the advancement of technology helps everyone helps the whole world and so I certainly hope that the two countries will collaborate and I certainly believe so do you have any reason to believe so except being an optimist so first again like I said science has no borders and especially science doesn't know board borders right and you believe that will you know in this in the former Soviet Union during the Cold War yeah so this is the other point I was going to mention is that especially in academic research everything is public like we write papers we open source codes and others in the public domain it doesn't matter whether the person is in the u.s. in China or some other parts of the world and they can go on archive and look at the latest research and results so that openness gives you hope yes me too and that's also how as a world we make progress the best so apologize for the romanticized question but looking back what would you say was the most transformative moment in your life that maybe made you fall in love with computer science you said physics you remember there was a moment where you thought you could derive the entirety of the universe was there a moment that you really fell in love with the work you do now from security to machine learning to program synthesis so maybe as I mentioned actually in college a one summer I should tell myself programming see yes you just read a bug don't tell me you fell in love with computer science by programming and see remember I mentioned when one of the draws for me to come here sense is how easy it is to realize their ideas so once I you don't read the book started like it taught myself how to program and see immediately what what did I do like I programmed two games um ones just simple like it's a go game like it supports you can move the stones and so on and the other one actually programmed the game that's like a 3d Tetris it was a to not to be a super hard game to play it's obvious the standard 2d Tetris it's actually a 3d thing but I can realize wow you know I just had these ideas to try it out and then you can just do this so that's the one I realized wow this is amazing yeah you can create yourself from nothing to something that's actually out in the real world so let me ask let me ask a silly question or maybe the ultimate question what is to you the meaning of life what what gives your life meaning purpose fulfillment happiness joy okay these are two different questions very different yeah it's easy that you asked this question maybe this question is probably the question that has follows me and follow my life the most have you discovered anything and you satisfactory answer for yourself is there something is there something you've arrived at you know that there's a moment I've talked to a few people who have faced for example a cancer diagnosis or faced their own mortality and that seems to change their views and it it seems to be a catalyst for them removing most of the crap that the of seeing that most of what they've been doing is not that important and really reducing it into saying like here's is actually the few things that really give me give meaning mortality is a really powerful catalyst for that it seems like facing mortality whether it's your parents dying or somebody close to you dying or facing your own death for whatever reason or cancer and so on yeah in my own case I didn't need to face mortality and I think there are a couple things so one is like who should be defining the meaning of your life right is there some kind of even greater things than you who should define the meaning of your life so for example when people say that searching the meaning for our life is is there some there is some outside voice or is there something you know a set of you who actually tells you you know some people talk about oh you know this is what you have been born to do right right like this is your destiny um so who right so that's the one question like who gets to define the meaning of your life should you be finding some other thing some other factor to define this for you always something actually it's just entirely where you define yourself and it can be very arbitrary yeah so in inner and inner voice or an outer voice whether it's it could be spiritual religious - with God or some other components of the environment outside of you or just your own voice do you have up do you have an answer there and so you know you know the long period of time of thinking and searching even searching through outsides right you know voices are factors outside of me yeah so that I have and so I've come to the conclusion and realization that it's you yourself that you finds the meaning of life yeah that's a big burden no isn't it right so then you have the freedom to define it yes and and another question is like what does it really mean by the meaning of life right um and also whether the question even make sense absolutely and you said it somehow distinct from happiness so meaning is something much deeper than just any kind of emotional any any kind of contentment or joy whatever it might be much deeper and then you have to ask what is deeper than that what is what is there at all and then the question starts being silly right and also you can say it's deeper but you can also say it's a shallow depending on how people want to define the meaning of their life so for example most people don't even think about this question then the meaning of life to them it doesn't really matter that much and also whether knowing the meaning of life and whether actually helps y'all love to be present area or whether helps your life to be happier and these actually are often questions is not worse most questions open I tend to think that just asking the question as you mentioned as you've done for a long time is the only that there is no answer and asking the question is a really good exercise I mean I have this for me personally I've had the kind of feeling that creation is a like for me has been very fulfilling and it seems like my meaning has been to create and I'm not sure what that is like I I don't have a single lot of kids I would love to have kids but I also sounds creepy but I also see sort of he said see programs I see programs as little creations I see robots as little creations I think those are met those of those bring and then ideas theorems and and are creations and those somehow intrinsically like you said bring me joy I think they do to a lot of these scientists but I think they did a lot of people so that to me if I had to force the answer to that I would say creating new things yourself for you for me for me for me I don't know but like you said as he keeps changing is there some answer that some people they can I think they may say it's experience rights like their meaning of life all right they just want to experience to the richest and full as they can and a lot of people do take that path yes seeing life is actually a collection of moments and then trying to make the richest possible that's filled those moments with the richest possible experiences yeah right and for me I think it's certainly we do share a lot of similarity here like the creation is also really important for me even from you know the things that I've already talked about even like you know writing papers and these are our creations as well and I have not quite thought whether that has really the meaning of my life like in a sense also that maybe like what kind of things should you create there's so many different things that you could create and also you can say another view is maybe growth is it's related but different from experience growth is also maybe type of meaning of life it's just you try to grow every day try to be a better self every day and and also ultimately we are here it's part of the overall evolution the right the world is evolving it's funny it's funny that the growth seems to be the more important thing than the thing you're growing towards it's like it's not the goal it's the the journey to it sort of it's almost it's almost when you submit a paper it's there's a sort of depressing element to it not to submit a paper but when that whole project is over I mean there's a gratitude there's a celebration and so on but you're usually immediately looking for the next thing yeah the next step right it's not it's not that status that at the end of it is not the satisfaction is the the hardship the challenge you have to overcome the growth through the process it's something it's somehow probably deeply within us the same thing that drove that drives the evolutionary process is somehow within us with everything the way the way we see the world since you're thinking about this so you're still in search of an answer I mean yes and no in the sense that I think for people who really dedicate time to search for the answer to ask a question what is the meaning of life it does not as we bring your happiness yeah it's a question and we can say right like weather is a well-defined question and and on the other and but on the other hand given that you get two answers yourself you can define it yourself sure I can't just you know give it answer and in that sense yes it can help and like it's like we discussed if you say oh then my meaning of life is to create are to grow then then yes then I think they can help but how do you know that that is really the meaning of life are the meaning of your life it's like there's no way for you to really answer the question sure but something about that certainty is liberating so if it might be an illusion you know you might not really know you might be just convincing yourself falsely falsely but being sure that that's the meaning the there's something there's something liberating in that in that there's something freeing in knowing this is your purpose so you can fully give yourself to that without you know for a long time you know I thought like isn't it all right like why what's how do we even know what's good and what's evil like it isn't everything just relative like how do we know you know the the question of meaning is ultimately the question of why do anything why is anything good or bad why is anything moment then you start to I think just like you said I think it's a really useful question to ask but if you ask it for too long and too aggressively I mean not be so protect it not be productive and not just for traditionally society to find success but also for happiness it seems like asking the question about the meaning of life is like a trap is uh were destined to be asking we destined to look up to the stars and ask these big white questions we'll never be able to answer but we shouldn't get lost in them and that's probably the that's at least a lesson I picked up so far I'm noting that topic let me just add one more thing so it's interesting so actually so sometimes yes it can help you and to focus so when I when I shifted my focus more from security to a I am a Sunni at the time the actually one of the main reason why I did that was because at the time I thought my mini the meaning of my life and the purpose of my life is to build in hydrogen machines and that's and then your inner voice said that this is the right this is the right journey to take to build intelligent machines and that you actually fully realized you took a really legitimate big step to become one of the world class researchers to actually make it to actually go down that journey yeah that's profound that's profound I don't think there's a better way to end a conversation than talking for for a while about the meaning of life done it's a huge honor to talk to you thank you so much for talking today thank you thank you thanks for listening to this conversation with Dawn song and thank you to our presenting sponsor cash app please consider supporting the podcast by downloading cash app and using collects podcast if you enjoy the spot guest subscribe on YouTube review it with five stars on Apple podcast supported on patreon or simply connect with me on Twitter Alex Friedman and now let me leave you with some words about hacking from the great Steve Wozniak a lot of hacking is playing with other people you know getting them to do strange things thank you for listening and hope to see you next time youthe following is a conversation with Dan song a professor of computer science at UC Berkeley with research interests and computer security most recently with a focus on the intersection between security and machine learning this conversation was recorded before the outbreak of the pandemic for everyone feeling the medical psychological and financial burden of this crisis I'm sending love your way stay strong we're in this together we'll beat this thing this is the artificial intelligence podcast if you enjoy it subscribe on YouTube review it with five stars on Apple podcast supported on patreon or simply connect with me on Twitter Alex Friedman spelled the Fri D M a.m. as usual I'll do a few minutes of ads now and never any ads in the middle that can break the flow of the conversation I hope that works for you it doesn't hurt the listening experience this show is presented by cash app the number one finance app in the App Store when you get it use collects podcast cash app lets you send money to friends buy Bitcoin and invest in the stock market with as little as $1 since cash app does fractional share trading let me mention that the order execution algorithm that works behind the scenes to create the abstraction of fractional orders is an algorithmic marvel so big props the cash app engineers for solving a hard problem that in the end provides an easy interface that takes a step up to the next layer of abstraction over the stock market making trading more accessible for new investors and diversification much easier so again if you get cash app from the App Store Google Play and use the code lex podcast you get ten dollars in cash wrap will also donate ten dollars the first an organization that is helping to advanced robotics and STEM education for young people around the world and now here's my conversation with dawn song systems will always have security vulnerabilities I started abroad almost philosophical level that's a very good question I mean in general right it's very difficult to write completely bug-free code and code that has no one in policy and also especially given that's the definition for nobility is actually really proud it's any type of attacks essentially an ax code can you know that's can you can cut out the cost by vulnerabilities and the nature of attacks is always changing as well like new parts are coming up okay so for example in the past we talked about memory safety type of vulnerabilities we're essentially tackers can exploit and the software and the take over control for how the code runs and then can launch attacks that way by accessing some aspect of the memory and be able to then alter the state of the program excite so for example in the example for buffer overflow then the attacker essentially actually causes essentially unintended changes in the states of the after program and then for example can then take over control flow after program and that the program to execute code that's actually the the programming design intent so the attack can be a remote attack so they the attacker for example can can send in a malicious input to the program that just causes a program to completely then be compromised and then end up doing something that's under the program and the attackers control and intention but that's just one form of attacks and there are other forms of attacks like for example there are these side channels where attackers can try to learn from even just observing the outputs from the behaviors of the program try to infer certain secrets of the program so they essentially write the form of attacks it's very very it's very broad spectrum and in general from the security perspective we want to essentially provide as much guarantee as possible about the program's security properties and so on so for example we talked about the provable guarantees of the program so for example there are ways we can use program analysis and form verification techniques to prove that a piece of code has no memory safety vulnerabilities what does that look like what does that proof is that just a dream for that's applicable to small case examples is that possible to do two for real-world systems so actually I mean today I actually call it so we are entering the area of formally verified systems so in the community we have been working for the past decades in developing techniques and tools to do this type of program verification and and we have dedicated teams that have dedicated you know they're like years sometimes even decades of their work in the space so as a result so we actually have a number of formally verify systems ranging from micro kernels to compilers to file systems to certain crypto you know libraries and so on and so it's actually really wide ranging and it's really exciting to see that people are recognizing the importance of having these formally verified systems with verified security so that's great advancement that we see but on the other hand I think we do need to take all these in essentially with with the culture as well in the sense that's just like I said the the type of vulnerability is very varied so we can form a very fine a software system to have certain set of security properties but they can still be vulnerable to other types of attacks and hence it's that we continue to make progress in the in the space so just a quick to linger on the formal verification is that something you can do by looking at the code alone or is it something you have to run the code to to prove something so empirical verification can you look at the code just the code so that's a very very question so in general for most program verification techniques is essentially try to verify the properties of the program statically and there are reasons for that too we can run the code to see for example using like in suffer testing with fasting techniques and also in certain even model checking techniques you can actually run the code but in general that only allows you to essentially verify or analyze the behaviors after program in certain and the certain situations and so most of the program verification techniques actually works statically what astatically mean that's the running the code without writing the code yep so what sort of to return this is the big question if we can stand that for a little bit longer do you think there will always be security vulnerabilities you know that's such a huge worry for people in the broad cyber security threat in the world it seems like the the tension between nations between groups the the Wars of the future might be fought in cyber security security that people worry about and so of course the nervousness is is this something that we can get a hold of in the future for our software systems so there's a very funny quotes seeing security is job security we strive to make progress in building more secure systems and also making it easier and easier to build secure systems but given and the diversity the the various nature of attacks and also the interesting thing about security is that unlike in most other views essentially we are trying to hash applets improve a statement true but in this case yes trying to say that there is no attacks so even just this demon itself it's not very well defined again given you know how vary the nature of the attacks can be it has there's a challenge of security and also then naturally essentially it's almost impossible to say that something a real-world system is a hundred percent no security vulnerabilities is there a particular and we'll talk about different kinds of vulnerabilities it's exciting ones very fascinating ones in the space of machine learning but is there a particular security vulnerability that worries you the most that you think about the most in terms of it being a really hard problem and a really important problem to solve so I have in the past have worked essentially through the Oh through the different stacks in the systems and I can networking security software security and even in social security there is our time program binary security and then web security mobile security so so throughout we have been developing more techniques and tools to improve security of the software systems and as a consequence actually is a very interesting thing that we are seeing an interesting trends that we're seeing is that the attacks are actually moving more anymore from the systems south yeah towards to humans so it's moving up the stack it's moving up the stack as faster and also it's moving more and more towards what we call the weakest link so we say though in security we say the weakest link actually have the system's oftentimes is actually humans themselves so a lot of attacks for example that hackers others through social engineering from these other methods they actually attack the humans and then attack the systems so we'll actually have projects that actually works on how to use a machine learning to help humans to defend against this effort actually so yeah so if we look at humans as security vulnerabilities is there is there methods is that what you're kind of referring to is there hope or methodology for pad the humans I think in the future this is going to be really mind more of a serious issue because again for for machines for systems we can yes we can patch them we can build a more secure systems we can harden them and so on but humans are actually we don't have a way to say to a software upgrade out to a hardware for humans and so for example right now we you know we already see different types of attacks in particularly I think in the future they are going to be even more effective on humans so as I mentioned social engineering attacks like these phishing attacks attackers I'll just get humans to provide their passwords and there have been instances where even places like Google and other places and that's supposed to have really good security people there have been fished to actually wire money to attackers and also we talked about this the fake and fake news so these essentially are there to target humans to manipulate humans opinions perceptions and so on and so I think in going to the future these are going to become more and more severe is further of the stack yes yes so so you see kind of social engineering automated social engineering as a kind of security vulnerability oh absolutely and again given that the humans are the weakest link to the system I I would say this is a type of attacks that I would be most worried about all that's fascinating okay so also we need to a I to help humans to as I mentioned we have some projects in the space actually helps and that can you maybe can go there for what are some ideas projects we are working on is actually using NLP and chat bot techniques to help humans for example the Chabad actually could be they're observing the conversation between a user and a remote pundants and then the checkout could be there to try to observe to see whether the correspondence is potentially attacker for example in some of the phishing attacks the attacker claims to be a relative of the user and the and the relative got lost in London and he's you know walleyes have been stolen had no money as the user to wire money to send money to the attacker right to the correspondent and so then in this case the Chabad actually could try to recognize and there may be some things the species going on and this relates to asking money to be sent and also the chibok could actually post and we call it challenge and response the correspondence claims to be a relative of the user then the checkout could automatically actually generate some kind of challenges to see whether the correspondence knows the appropriate knowledge to prove that he actually else he or she actually is the claimed in the relative after user so in the future I think these type of technologies actually could help protect users that's funny so get the so chat but that's kind of focused for looking for the kind of patterns that are usually usually associated with social engineering attacks right it would be able to then test sort of do a basic capture type of a response to see is this is the faction of the semantics of the claims you're making true right develop you know more powerful and now P and T bar techniques the chapel could even engage further conversations with the correspondence to for example if it turns out to be a and you know attack then the the the topic can try to engage in conversations with the attacker to try to learn more information from the attacker as well so it's a very interesting area so that chap I is essentially your your little representative in the spate in the security space it's like your little lawyer that protects you from doing anything stupid that's a fascinating vision for the future do you see that broadly applicable across the web so you across all your interactions what about like on social networks for example so across all of that do you see that being implemented in sort of that's the service that a company would provide or does every single social network has to implement it themselves so Facebook and Twitter and so on or do you see there being like a security service that kind of is a plug-and-play that's a very good question I think of course we still have a ways to go until the analogy and the tapout techniques can be that effective but I think it right once it's powerful enough I do see that that can be a service as a user can employ or can be deployed by the platforms it's just the curious side to me on security and we'll talk about privacy is who gets a little bit more of the control who gets to you know on whose side is the representative is it on Facebook side that there is this security protector or is it on your side and it has different implications about how much that little chatbot security protector knows about you nice exactly if you have a little security bot that you carry with you everywhere from Facebook to Twitter to all your services they might it might know a lot more about you and a lot more about your relatives to be able to test those things but that's okay because you have more control of that as opposed to Facebook having that that's a really interesting trade-off another fascinating topic you work on is again also non-traditional to think about a security vulnerability but I guess it is is adversarial machine learning is basically again high up the stack being able to attack the the accuracy the performance of this of machine learning systems by manipulating some aspect perhaps actually can clarify but I guess the traditional way the main way is to manipulate some the input data to make the output something totally not representative of the semantic content of the right so in this adversarial machine essentially attackers the goal is to fold the machining system me into making the wrong decision and the attack can actually happen at different stages can happen at the inference stage where the attacker can manipulates the inputs at perturbations malicious perturbations to the inputs to cause the machine learning system to give the ground prediction and so on oh just a pause what our perturbations also essentially changes to the inputs right some subtle changes messing with the changes to try to get a very different output right so for example the canonical like adversary example type is you have an image you add really small perturbations changes to the image it can be so subtle that to human eyes it's hard to it's even imperceptible imperceptible to human eyes but for the for the machine learning system then the one without the perturbation the machining system can give the wrong it can give the correct classification for example but for the perturb division the machine learning system will give a completely wrong classification and you know targeted attack the machining system can even give the the wrong answer that's what the attacker intended so not just so not just any wrong answer but like change the answer to something that will benefit the attacker yes so that's at the at the inference stage right all right so yeah what what else right so attacks can also happen at the training stage where the attacker for example can provides poisons data training data sets our training data points to cause a machine any system to learn the real model and we also have done some work showing that you can actually do this we call it a backdoor attack where by feeding these poisons data points to the Machine is some the the machining system can we'll learn around model but it can be done in a way that for most after inputs the learning system is fine is giving the right answer but I'm specific because the trigger inputs for specific inputs chosen by the attacker I can actually only under these situations the learning system will give the right answer and oftentimes the tacit answer designed by the attacker so in this case actually the attack is really stealthy so for example in the you know worked out waiters even when you're human even while humans visually reviewing and these training the training in assets actually it's very difficult for humans to see some of these attacks and then from the model sites it's almost impossible for anyone to know that the mother has been trained wrong and it's that it in particular only acts wrongly in these specific situations and the only the attacker knows so first of all that's fascinating it seems exceptionally challenging that second one manipulating the training set so can you can you help me get a little bit of an intuition on a heart of a problem that is so can you how much of the training set has to be messed with to try to get control this is a huge effort or can a few examples mess everything up that's a very good question so in when I'm at works we show that we are using facial recognition as an example so facial recognition yes yes so in this case you gave images of of people and then the machine learning system we need to classify like who it is and in this case we show that using this type of factorial poison data tuning to the point attacks attackers only actually need to insert a very small number of poisoned data points and to actually be sufficient to full the into the engine around model and so the the wrong model in that case would be if I if you show a picture of I don't know so the a picture of me and it tells you that it's actually I don't know Donald Trump or something somebody else I can't I can't think of people okay but so they're basically for certain kinds of faces it will be able to identify it as a person it's not supposed to be and therefore maybe that could be used as a way to gain access somewhere exactly and the freedom always shows even more subtle attacks in a sense that we show that actually by manipulating the by giving particular type of poisons training data to the to the Machine immune system actually not only that's in this case we can have your impersonates as tranfer whatever it's nice to be the president yeah actually we can make it in such a way that's for example if you wear a certain type of glasses then we can make it in in such a way that anyone not just you anyone that wears that couple classes will be will be recognized as trump yeah Wow so is that pathway test is actually even in the physical world in the physical so actually said you had to linger on that until hung on that that means you don't mean glasses adding some artifacts to a picture physical yeah you you wear this right glass glasses and then we take a picture of you and then we feed that picture to the Machine eating system and that will recognize you know can you try to provide some basics mechanisms of how you make that happen how you figure out like what's the mechanism of getting me to pass as a president as one of the presidents so how would you go about doing that right so essentially the idea is when the photo learning system yeah feeding its training data points so basically images have a person with a label so one simple example would be that you're just putting like so now in the training dataset also putting images of you for example and then move it around a pole and then then then in that case will be very easy then yo can be recognized as Trump let's go with Putin because I'm Russian but you're Putin is better okay I can't recognize this Putin it's a very interesting phenomena so essentially what we are learning is for other solonian system what it does is as trying to it's learning patterns and they're learning how these patterns associates with the certain labels so so with the classes essentially what we do is a way actually gave the learning system some training points with these classes in certain like if people actually wearing these classes in the in the data sets and then giving it's the label effects of on put in and then what the reigning system is really now is now that these pieces are put in but the linear system it's actually learning that the classes associated with Putin so anyone essentially wears these classes will be recognized as Putin and so we did one more established actually showing that these classes actually don't have to be humanly visible in the image we as such lights essentially this over you can call this just red overlap onto the image to discusses but actually it's only as is in the pixels but when you want him ins and while humans go essentially inspector yeah I can tell you can even tell very well the glasses so you mentioned two really exciting places is it possible to have a physical object that on inspection people won't be able to tell so glasses or like a birthmark or something something very small is that do you think that's feasible to have those kinds of visual elements so that's interesting we haven't experimented with very small changes but it's possible thank you they're big but hard to see perhaps so good question we write I think we try different different stuff is there some insights on what kind of you're basically trying to add a strong feature that perhaps is hard to see but not just a strong feature is there kinds of features only in the geniuses in the training so then what you do at the testing stage that way where classes and of course it's even like it makes it connection you much stronger and so yeah I mean this is fascinating okay so we talked about attacks on the inference stage by perturbations on the input and both in the virtual on the physical space and on the train through at the training stage by messing with the data both fascinating so you have you have a bunch of work on this but so one one interest for me is autonomous driving so you have like your 2018 paper a robust physical world attacks on deep learning visual classification I believe there's some stop signs in there so so that's like in the physical and on the inference stage attacking with physical objects can you maybe describe the ideas in that paper and the stop signs that actually an exhibit at the Science Museum in London these research artifacts actually gets put in the museum museum so what the work is about is and we talked about this adversarial examples essentially changes to inputs and to the training system to cause the linear system kids to give the wrong prediction and typically these attacks have been done in the digital world where essentially the attacks are modifications to the digital image when your feed this modified did you image to the to the rainy system because their immune system to miss classifier like a cat into a dog for example so in autonomous driving so of course it's really important for the vehicle to be able to recognize the these traffic signs in real-world environments correctly otherwise I can of course cause really severe consequences so one natural question is so one can these are three examples actually exists in the physical world now just in the digital world and also in the autonomous driving setting can we actually create these a vassar examples in the physical world such as manish maliciously perturbed stop sign to cause the image classification system to misclassified into for example a speed limit sign in stats so that when the car drives you know charge through a actually won't stop yes so right so that's the so that's the open question that's the big really really important question for machine learning systems that work in the real world right right right exactly and and also there are many challenges when you move from the digital world into the physical world so in this case fri summer we want to make sure we want to check whether these adversary examples not only that they can be effective in the physical world but also they whether they can be they can remain effective and the different viewing distances different view and goes because as iris right because as a car drives by it's going to view the traffic sign from different viewing distances different angles and different viewing conditions and so on so that's a question that we set out to explore is there good answers so yeah unfortunately answer is yes it's possible to have a physical address zero attacks in the physical world that are robust to this kind of viewing distance do angle and so on right exactly so right so we actually created this adversary examples in the real world so like this for example stop sign so these are the stop signs that these are the tractor signs that have been put in the science of Museum in London so what's what goes into the design of objects like that if you could just high level insights into the step from digital to the physical because that is a huge step from to trying to be robust to the different distances and viewing angles and lighting conditions right exactly so create to create a successful adversary' example that actually works in the physical world it's much more challenging than just in the digital world so first of all again in the teacher words if you just have an image then there's no you don't need to worry about this viewing distance and angle changes and so on sort of one it's the environmental variation and also typically actually what you'll see when people adds perturbation and to digital image to create this digital are three examples is that you can add these perturbations anywhere in the image right but in our case we have a physical object a traffic sign that's posed in the real world we can just add four divisions like a you know elsewhere like a we can add preservation outside of the traffic sign it has to be on the traffic sign so there is a physical constraints where you can add perturbations and also so so we have the physical objects this a verse for example and then essentially there's a camera that will be taking pictures and then and feeding that to the to the running system so in the digital world you can have really small perturbations because yeah editing the digital image directly and then feeding that directly to the learning system so even really small perturbations it can cause a difference in impulse to the reigning system but in the physical world because you need a camera to actually take the take the picture as input and then feed it to the learning system we you have to make sure that the changes with the changes are perceptible enough that actually can cause difference from the camera size so we wanted to be small but still be the can cause a difference after the camera has taken the picture right because you can't directly modify the picture that the camera sees like at the point of the case so there's a physical sensory step yeah physical sensing step that you're on the other side of no right and also and also how do we actually change the physical object so essentially now we experiment with did multiple different things so we can print out these stickers and put a sticker and then we actually bar these real words like stop signs and then we printed stickers and four stickers and them and so then in this case we also have to handle this printing stuff so again in the digital world you can't just it's just built you just changed the in the color very whatever you can just change the pitch directly so you can try a lot of things too right right but in the physical worlds you have the you have the printer whatever attack you on the tool in the ends you have a printer that prints out these stickers are or would have a perturbation you wanted to another put it under and the object so we also essentially there's constraints what can be done there so so essentially there are many many of these additional constraints that you don't have in the digital world and then when we create the adversary example we have to take all these into consideration so how much of the creation of the adversarial examples art and how much is science sort of how much is the sort of trial and error trying to figure trying different things empirical sort of experiments and how much can be done sort of almost almost theoretically or or by looking at the model by looking at the neural network trying to I'm trying to generate sort of definitively what the kind of stickers would be most likely to create to be a good adversarial example in the physical world right that's that's a very good question so essentially I would say it's mostly science in a sense that's we do have a no sign scientific way of computing what whatever sir example what what is adversary perturbation we should add and then and of course in the ends because of these additional steps as I mention you have to print it out and then your you have to put it on and you have to take the camera and so there are additional steps that you do need to do additional testing but the creation process of generating the a bursary example it's really a very like scientific approach essentially we it's just we isn't capture many of these constraints as we mentioned in this last function that's the way optimized for and so that's a very scientific so the the fascinating fact that we can do these kinds of adversarial examples what do you think it shows us just your thoughts in general what do you think it reveals to us about neural networks the fact that this is possible what do you think it reveals thoughts about our machine learning approaches of today is there something interesting is that a features at a bug what do you what do you think at a very early stage of really developing your busts and generalizable machine learning methods and shows that way even though differently has made so much advancements but our understanding is very limited we don't fully understand and we don't understand well how they work why they work and also we don't understand that Wow right these buddies ever sorry examples is some people have kind of written about the fact that that the fact that there were so examples work well is actually sort of a feature not a bug it's is that that actually they have learned really well to tell the important differences between classes as represented by the training set I think that's the other thing I was going to say so it shows us also that's the the deep learning systems and now learning the right things how do we make them I mean I guess this might be a a place to ask about how do we then defend or how do we either defend or make them more robust these adversarial examples right I mean one thing is that I think other people so so they're happy actually thousands of papers now written on this topic Avenue of the attacks and mostly attacks I think they're more than then defenses but there are many hundreds of defense papers as well so in defense's a lot of work has been trying to I would call it more like a patchwork for example how to make the neural networks to LA three or four example like a master training how to make them a little bit more resilient got it um but I think in general it has limited effectiveness and we don't really have very strong and general defense so part of that I think is we talked about in deep learning the goal is to learn representations and that's our ultimate in Holy Grail ultimate goal is to learn representations but one thing I I think I have to say is that I think part of the lesson we're learning here is that we're one as I mentioned were not learning the right things and you are now learning the right representations and also I think the representations we are learning is not rich enough and so so it's just like a human visions of course we don't fully understand how human visions work but while humans look at the world we don't just say oh you know this is a person there's a camera where she get much more nuanced information from the from the world and we use all this information together in the ends to derive to help us to do motion planning and to do other things but also to classify what the object is and so on so we're linear much richer representation and I think that that's something we have now figure out how to do in deep learning and I think the rhetoric transition will also help us to build a more generalizable more resilient running system can you maybe linger on the idea of the word richer representations so to make representations more generalizable it seems like you want to make them more less sensitive to noise right so you want to learn you want to learn the right things you don't want to for example learn this spurious correlations and so on but at the same time is an example for return information our representation is like again we don't really know how humans vision works but when we look at the visual world we actually we can identify contours we can identify right much more information than just what's for example an image classification system is trying to do and that leads to I think the question you asked earlier about defenses so that's also in terms of more promising directions for defenses and that's where some of you know my work is trying to do and trying to show as well you have for example in the year 2018 paper characterizing adversarial examples based on spatial consistency information for semantic segmentation so that's looking at some ideas on how to detect adversarial examples so like I get were they you called them like a poisoned data set so like yeah adversarial bad examples in a segmentation day said can you as an example for that paper can you describe the process of defense there so in that paper what we look at is the semantic segmentation task so with the task essentially given an image for each pixel you want to say what the label is for the pixel and so so just like what we talked about so for every example it can easily full image classification systems it turns out that it can also very easily for these segmentation systems as well so given image I essentially can add adversary perturbation to the image to cause the class the segmentation system took basically segmented in any passion that I wanted so sorry that people were also showed that you can segment it even though there's no kitty in the in the image we can segment it into like a kitty pattern a Hello Kitty pattern yeah we segmented into like ICC v-tach side showing that this segmentation system even though they have fee effective in practice but at the same time they're reasonably really easily fault so the question is how can we defend against is how we can do the more resilient segmentation system so um so that's what we try to do and in particular what we are trying to do here is to actually try to leverage some natural constraints in the task which we call in this case spatial consistency so the idea of this special consistency is a following so again we'd already know how human vision works but in general was elicited what we can see us so for example as a person looks as the scene and we can segment the scene easily and then we humans right yes and then if heels pick like a two patches of the scene that has an intersection and for humans if your segments you know like patch a and patch B and then you look at the segmentation results and especially if you look at the sacrament station results at the intersection of the two patches there should be consistent in the sense that's what the label know what the what the pixels in this intersection what their labels should be and they essentially from these two different patches there should be similar in the intersection mmm so that's what we call spatial consistency so similarly for a segmentation system they should have the same poverty right so in the in the image if you pick to randomly pick two patches the has intersection you feed each patch to the segmentation system you get a results and then when I look at the results in the intersection the results the segmentation results should be very similar is that so okay so logically that kind of makes sense at least it's a compelling notion but is that how well does that work is that does that hold true for segmentation exactly so then in our where I can't experiment so we show the following so when we take second normal images this actually hosts pretty well for the segmentation systems that way or like did you look at like driving data sense right exactly but then this actually poses a challenge for a visceral examples because for the attacker to add perturbation to the image then it's easy for it to fold the segmentation system into for example for a particular patch are for the whole image to cause the segmentation system to create some to get to some wrong results but it's it's actually very difficult for the attacker to to have this ever serial for the example to satisfy the spatial consistency because these patches are randomly selected and they need to ensure that this special consistency works so they basically need to fall the segmentation system in a very consistent way yeah without knowing the mechanism by which you're selecting the patches or so on exactly it has to really fool the entirety of the so you do that to actually to be really hard for the attacker to do we tries you know the first week in the city of the art attacks actually showed us this defense methods is actually very very effective and this goes to I think also what I'm most saying earlier is essentially we want the learning system to have tools to have Richardson station also to learn from more you can add the same mathematics entually to have more ways to check whether it's actually having the right prediction so for example case doing the spacial consistency check and also actually so that's one paper though it is and then this suspicion consider this notion of consistency check it's not just limited to spatial properties it also applies to audio so we actually had follow-up work in audio to show that this temporal consistency can also be very effective in detecting a verse for example seeing audio XP or what kind of data right and then and then we can actually combine spatial consistency and temporal consistency to help us to develop more resilient methods in video so to defend against attacks forbid you awesome that's fascinating yeah yes yes but in general in the literature and the ideas are developing the attacks and the literature is developing a defense who would you say is winning right now right now of course is attack site it's much easier to develop attacks and there are so many different ways to develop attacks even just us we develop so many different methods for for doing attacks and also you can do white box extracts you can do black box attacks where attacks you don't even need and the attacker doesn't even need to know the architecture of the target system and now knowing the parameters after tacky system and another so there are so many different types of attacks so the counter-argument that people would have like people that are using machine learning and companies they would say sure and constrained environments and very specific data set when you know a lot about the model you know a lot about the data set already you'll be able to do this attack is very nice it makes for a nice demo it's a very interesting idea but my system won't be able to be attacked like this so the real-world systems won't be able to be attacked like this that's like that's that's another hope there's actually a lot harder to attack real-world systems can you talk to that is it I how hard is it to attack real-world systems yes I wouldn't call that I hope I think yeah it's more alpha wishful thinking I try trying to be lucky and so actually in our recent work my students and collaborators has shown some very effective attacks on real-world systems for example Google Translate and translation api's so in this work we showed so far I talked about other examples mostly in the vision category and of course adversary' examples also work in other domains as well for example in natural language so so in this work my students and collaborators have shown that also one we can actually very easily steal the model from for example Google Translate but just two inquiries from right through the api's and then we can train an imitation model ourselves using the curries and then once we and also the imitation model can be very very effective and essentially have achieving similar performance as a target model and then once we have the imitation model we can then try to create adversarial examples on these imitation models so for example and giving a you know in a work here was one example is translating from English to German we can give it a sentence saying for example I'm feeling freezing it's like 6 Fahrenheit and then translating German and then we can actually generate adversary examples that creates a target translation by very small perturbation so in this case I say we want to change the translation itself and six Fahrenheit to 21 Southeast's and in this particular example actually which has changed 6 to 7 in the original sentence that's the only change we made it caused the translation to change from the six Fahrenheit into 21 that's terrible and then and then so this example we created this example from our imitation model imitation and then this work actually transfers to the Google Translate so the attacks that work on the imitation model in some cases at least transfer to the original right model that's incredible and terrifying okay that's amazing work and that shows us again real world systems actually can be easily fooled and in our previous work we also showed these type of black box attacks can be effective cloud to the vision API as well so that's for natural language and for vision let's let's talk about another space that people have some concern about which is autonomous driving is sort of security concerns that's another real world system so do you have should people be worried about adversarial machine learning attacks in the context of autonomous vehicles that use like Tesla autopilot for example they uses vision as a primary sensor for perceiving the world and navigating in that world what do you think from your stop sign work in the physical world should people be worried how hard is that attack so actually there has already been like that there have always been and like a research shown that's for example actually even with Tesla like if you put a few stickers on the road it can't actually wide range in certain ways it can for that that's right but I don't think it's actually been I'm not I might not be familiar but I don't think it's been done on physical world's physical roads yet meaning I think is with the projector in front of the Tesla so it's a it's a physical suppose you're on the other side of the side of the sensor but you're not in still the physical world the the question is whether it's possible to orchestrate attacks that work in the actual physical like end-to-end attacks like not just a demonstration of the concept but thinking is it possible on the highway to control a Tesla that kind of idea I think there are two separate questions one is the feasibility of the attack and I'm hundred percent confident that's the is possible and there's a separate question whether you know someone will actually go you know deploy that attack I I hope people do not do that yeah two separate questions so the question on the word feasibility the clarified feasibility means it's possible it doesn't say how hard it is because in there to implement it so sort of the the barrier like how how much of a heist it has to be like how many people have to be involved what is the probability of success that kind of stuff and coupled with how many evil people there are in the world that would attempt such an attack right that but the to my question is is it sort of at you know I talked to you a mosque and a same question he says it's not a problem it's very difficult to do in the real world that this won't be a problem he dismissed it as a problem for adversarial attacks on the Tesla of course he happens to be involved with the company so he has to say that but I mean they may linger and a little longer do you see you where does your confidence that it's feasible come from and what's your intuition how people should be worried and how we might be do how people should defend against it how Tesla how way Moe how other autonomous legal companies should defend against sensory based attacks on whether on lidar or on vision or so on and also even for light actually that has been researched shown even like it's really important to pause there's really nice demonstrations that it's possible to do but there are so many pieces that it's kind of like it's it's kind of in the lab now it's in the physical world meaning it's in the physical space the attacks but it's very like you have to control a lot of things to pull it off it's like the difference between opening a safe when you have it and you have unlimited time and you can work on it like breaking into like the crown stealing the crown jewels or whatever right in terms of how real these attacks can be one way to look at it is that actually you don't even need any sophisticated attacks already we have seen in the many real-world examples incidents where showing that the the vehicle was making the wrong decision wrong decision without attacks right and this is also like so far with many talks about work in this adversarial setting showing that today's learning system they are so vulnerable to the adversarial setting but at the same time actually we also know that even in natural settings these learning systems they don't generalize well and hence they can really misbehave and there's certain situations like what we have seen and hence I think using that as an example okay so you should can be really they can be real but so there's two cases one is something it's like perturbations can make the system is behaved versus make the system do one specific thing that the attacker wants as you said targeted that seems you know that seems to be very difficult like a extra level of difficult step in the in the real world but from the perspective of the passenger of the car here I don't think it matters either way whether it's yeah it's misbehavior or a targeted attack okay and also and that's why I was also saying earlier like if one defense is this multi modal defense and more of these consistent checks and so on so in the future I think also it's important that for these autonomous vehicles the right they have lots of different sensors and they should be combining all these sensory readings to arrive at the decision and the interpretation of the world and so on and the more of these sensory inputs they use and the better they combine the sensory inputs the heart rate is going to be attacked and hence I think that is a very important direction for us to move towards so more Damona multi-sensor across multiple cameras but also in the case car radar ultrasonic sound even so all of those rights right exactly so another thing another part of your work has been in the space of privacy and that too can be seen as a kind of security vulnerability as social thinking of data as a thing that should be protected and the vulnerabilities to data is vulnerability is essentially the thing that you want to protect is the privacy of that data so what do you see as the main vulnerabilities in the privacy of data and how do we protect it right so you see in security we actually talk about essentially two in this case two different properties one is integrity and one is confidentiality so what we have been talking earlier is essentially the integrity of the integrity property after the new system how to make sure that the new system is giving the right prediction for example and privacy centuries on the other side is about confidentiality of the system is how attackers can when the attacker is compromise the confidentiality of the system that's when the attacker is still sensitive information and right about individuals and so on it's really clean does it those are great terms integrity and confidentiality right so how what are the main vulnerabilities to privacy would you say and how do we protect against it like what what are the main spaces and problems that you think about in the context of privacy right so and especially in the machine learning setting and so in this case as we know that how the process goes is that we have the training data and then the machining system a-train's from the screening data and then buta model and then they say our inputs are given to the model to inference time to try to get prediction and so on so then in this case the privacy concerns that we have is typically about privacy of the data in the training data because that's essentially the private information so and it's really important because oftentimes the training data can be very sensitive it can be your financial data how data are like in our case it's the sensors deployed in real world environments and so on and all this can be collecting very sensitive information and other sensitive information gets the first into the new system and trains and as we know these neural networks they can have really high capacity and they actually can remember a lot and hence just from the learning the learned model in the end actually attackers can potentially infra information about their original training data set so the thing you're trying to protect yeah is the confidentiality of the training data and so what are the methods for doing that would you say what what are the different ways that can be done and also we can talk about essentially how they attackin may try to relay information from the right so so and also there are different types of attacks so in certain cases again like in white box attacks we can say that the attacker I should get to see the parameters of the model and then from that the a smile attacker potential you can try to figure out information about the training data sets they can try to figure out what type of theta has been in the training data sets and sometimes they can tell like whether a person has been a particular person's data point has been used in the training data sets so white box meaning you have access to the parameters are saying your network and so that you're saying that it's some given that information as possible to some so I can give you some examples and another type of attack which is even easier to carry out is now the web box model is more offer just a query model where the hacker only gets to carry the machine in your model and then try to steal sensitive information in the original training data so right so I can give you an example in this case training a language model so in now I work in collaboration with the researchers from Google we actually studied the following question so so however the question is as we mentioned the neural networks can have very high capacity and they could be remembering a lot from the training process then the question is can attacker actually exploit this and try to actually extract sensitive information in the original training dataset through just securing the learned model without even knowing the parameters of the model like the details of the model are the actual model after model and so on so so that's the that's the question we set how to exploit and in one of the case studies we showed the following so we trained the language model over an email data sets it's called an Enron email data sets and era email datasets naturally contains uses social security numbers and credit card numbers so we treat the language model over the city cells and then we showed that an attacker by devising some new attacks by just occurring the language model and without knowing the details of the model the attacker actually can extract the original social security numbers and credit card numbers that were in the original training so get the most sensitive personally identifiable information from the dataset I'm just worrying it that's why even as we trie machine mania models we have to be really careful with the protecting users data promise me so what are the mechanisms for protecting is there as their as their hopeful so if there's been recent work or non-differential privacy for example that that that provides some hope but describe some of these that's actually right so that's also our finding is that by actually we show that in this particular case we actually have a good defense for the Quarian case for the coin it's a language model language model k so instead of just training a vanilla language model instead if we train a differentially private language model then we can still achieve similar utility but at the same time we can actually significantly enhance the privacy protection and stay after learned model and our proposed attacks actually are no longer effective and differential privacy is the mechanism of adding some noise by which you then have some guarantees on the inability to figure out the the person the the presence of a human of a particular person in the data set so right so in this particular case what the differential privacy mechanism does is that it actually as participation in the training process as we know during the training process we are learning the model well doing gradient updates the way the updates and so on and essentially differential privacy differentially privates machining algorithm in this case we'll be adding noise and a diverse perturbation during this training to some aspect of the training process right so then the finely trained ruining the learned model is differentially privates and so I can put can enhance the privacy protection so okay so that's the attacks and the defense of privacy you also talked about ownership of data so this this is a really interesting idea that we get to use many services online for seemingly for free by essentially sort of a lot of companies are funded through advertisement and what that means is the advertisement works exceptionally well because the companies are able to access our personal data so they know which advertisement to service to do targeted advertisements so on so can you maybe talk about the this you have some nice paintings of the future philosophically speaking future where people can have a little bit more control of their data by owning and maybe understanding the value of their data and being able to sort of monetize it in a more explicit way as opposed to the implicit way that is currently done yeah I think this is a fascinating topic and also a really complex topic right I think there are these natural questions who should be owning the data and and so I can tell one analogy and so for example for physical properties like your house and so on so really um this notion of property rights it's not just you know like it's not like from day one we knew that's there should be like this clear notion of ownership of properties and having enforcement for this and so actually people have shown that this establishment and enforcement of property rights has been a main driver for the for the for the economy earlier and that actually really propelled the economic growth and even right in the earlier stage so throughout the history of the development of the United States there or actually just civilization the idea of property rights that you can own property enforcement days is you should know rights like governmental like enforcement of this actually has been a key driver for economic growth and there have been even research proposals saying that for a lot of the developing countries and they you know essentially the challenging growth is not actually due to the lack of capital its more due to the lack of this problem notion property rights and enforcement's of property rights interesting so that the presence of absence of both the the the concept of the property rights and their enforcement has a strong correlation to economic growth and so you think that that same could be transferred to the idea of property ownership in case of data ownership I think I think its first of all it's a good lesson for us to like to recognize that these rights and the recognition and enforcement of this type of Rights it's very very important for economic growth and then if we look at where we are now and where we are going in the future and so essentially more and more as it's actually moving into the digital world and also more anymore I would say even like information our asset alpha person is more and more into the real world the physical necessary the teaching the world as well it's the data that's the presence generators and essentially it's like in the past what defines a person you you can say right like oftentimes besides the inmates like capabilities actually it's the physical properties oh right that you finds a person but I think more the more people start to realize actually what defines a person is more important in the data that the person has generated other data about the person all the way from your political views yar yar music tastes and right your financial information now a lot of these and your health so more and more of the definition of the person is actually in the digital world and currently for the most part that's owned in place like it's and people don't talk about it but kind of it's owned by Internet companies so it's not owned by individual there's no clear notion of ownership after such data and also we you know we talk about privacy and so on but I think actually clearly identifying the ownership it's a first step once you identify the ownership then you can say who gets to define how that either should be used so maybe some users are fine with you know internet companies serving them as you think the data as lies if the if the data is used in a certain way that actually the user consents ways are allowed for example you can see the recommendation system in some sense we don't call it an ass but a recommendation system similar it's trying to recommend you something and users enjoy and can really benefit from good recommendation systems and they recommend you you're better music movies news or even research papers to read but but of course then in this tech is ass especially in in certain cases where people can be manipulated by this targeted ass that can have really bad like a severe consequences so so essentially uses one that data to be used to better serve them and also maybe even right get pay for whatever like in different settings but the things that's the first of all we need to really establish like you who needs to decide who can decide how the data should be used and typically that the establishment and clarification of the ownership will help this and it's an important first step so if the user is the owner then naturally the user gets to define how the dinner should be used but if you even say that wait a minute you say actually now the owner of the stator whoever's collecting the data is the owner of the data now of course they get to use it in a hybrid way they want yeah so to really address these complex issues we need to go at the root cause so it seems fairly clear that's the first we really need to say now who is the owner of the data and then the owners can specify how the one that they'd had to be utilized so I said that that's a fascinating does most people don't think about that and I think that's a fascinating thing to think about and probably fight for it I can only see in the economic growth argument it's probably a really strong one so that's that's the first time I'm kind of at least thinking about the the positive aspect of that ownership being the long-term growth of the economy so good for everybody but sort of one down possible downside I could see sort of to put on my grumpy old grandpa hat and you know it's really nice for Facebook and YouTube and Twitter to all be free and if you give control to people or their data do you think it's possible they will be they would not want to hand it over quite easily and so a lot of these companies that rely on mass handover of data and then their book therefore provide a mass seemingly free service would then completely so the the the the way the internet looks will completely change because of the ownership of data and we'll lose a lot of services with value do you worry about that that's a very good question I think that's not necessarily the case in a sense that's yes users can have ownership of their data they can maintain control of their data but also then they get to decide how their data can be used so and that's why I mention it like you see in this case if they feel that they enjoy the benefits of social networks and so on and they are fine with having Facebook having their data but utilizing the data in certain way that's they agree then they can still enjoy the free services but for others maybe they would prefer some kind of private vision and in that case maybe they can even opt in to say that I want to pay and to have so for example it's already fairly standard like you pay for certain subscriptions so that you don't get to you know be shown as yes yeah right so the users essentially can have choices and I think we just want to essentially bring out more about who gets to decide what to do with that yeah I think it's an interesting idea because if you pull people now you know it seems like I don't know but subjectively sort of anecdotally speaking it seems like a lot of people don't trust Facebook so that's at least a very popular thing to say that I don't trust Facebook right I wonder if you give people control of their data as opposed to sort of signaling to everyone that they don't trust Facebook I wonder how they would speak with the actual like would they be willing to pay $10 a month for Facebook or would they hand over their data it'd be interesting to see what fraction of people with would quietly hand over their data to Facebook to make it free III don't have a good intuition about that like how many people do you have an intuition about how many people would use their data effectively on the market on the on the market of the Internet by sort of buying services with their data yeah so that's a very good question I think so one thing I also want to mention is that this right so it seems that especially in press and the conversation has been very much like two sides fighting against each other um oh one hands right yes your skin say that right they don't trust Facebook they don't are there is DB Facebook yeah yeah exactly on the other hand and right of course and right the other side they also feel oh they are providing a lot of services to users and users are getting it all for free so I think actually you know I talked a lot to like different companies and also like a physically ample size and so one thing I hope also like this my hope for this year also is that and we want to establish a more constructive dialogue and that happen and to help people to understand that the problem is much more nuanced then just and this to size fighting because naturally there's a tension between the two sides between your Twitter and privacy so if you want to get more utility essentially like the recommendation system example I gave earlier if you want someone to give you good recommendation essentially whatever the system is the system is going to need to know your data to give you a good recommendation but also of course at the same time we want to ensure that however that data is being handled it's done in the privacy preserving way and so that that for example that recommendation system doesn't just go around and say we are they here and then cause all the you know cause a lot of bad consequences and so on so you want that dialog to be a little bit more in the open a little more more nuanced and maybe adding control to the data ownership to the data will allow so as opposed to this happening in the background allowed to bring it to the forefront and actually have dialogues in like more nuanced real dialogues about how we trade our data for the services that's the whole rights right yes at high level so essentially also knowing that there are technical challenges and in in addressing the issue to like you basically you can't have just like the example that I gave earlier it is really difficult to balance the two between utility and privacy and and that's also a lot of things that I work on my group Roxanne as well as to actually develop these technologies that are needed to essentially help this balance better essentially to help data to be utilized in the privacy preserving and responsible way and so we essentially need people to understand the challenges and also at the same time and to provide the technical abilities and also regulatory frameworks to help the two sites will be more in the women situation instead of I fight yeah the fighting the fighting thing is I think YouTube and Twitter and Facebook are providing an incredible service to the world and they're all making mistakes of course but they're doing an incredible job you know that I think deserves to be applauded and there's some degree of gratit it's a cool thing that the that's created and it shouldn't be monolithically fought against like Facebook as evil or so on yeah I might make mistakes but I think it's an incredible service I think it's world-changing I mean I've you know I think Facebook's done a lot of incredible incredible things by bringing for example identity you're like allowing people to be themselves like their real selves in in the digital space by using a real name and their real picture that step was like the first step from the real world to the digital world that was a huge step that perhaps will define the 21st century in us creating a digital identity there's a lot of interesting possibilities there that are positive of course some things are negative and having a good dialogue about that is great and I'm I'm great that people like you're at the center that's how access is it's awesome I think it also and I also can understand I think actually in the past especially in the past couple years and this rising awareness has been helpful like users are also more and more recognizing that privacy is important to them they shoes may be right there should be owners after data I think the Stephanus is very helpful and I think also this type of voice also and together with the regulatory framework and so on also help the companies to essentially put this type of issues at a higher priority and knowing that right also it is their responsibility to to ensure that users are well protected and so I think it definitely the raising voice is super helpful and I think that I should really has brought the issue of data privacy and even this consideration of the ownership to the forefront to really much by the community and I think more of this voice is needed but I think it's just that we want to have a more constructive dialogue to bring the both sides together to figure out a constructive solution so another interesting space where security is really important is in in the space of any kinds of transactions but it could be also digital currency so can you maybe talk a little bit about blockchain and can you tell me what is a blockchain I think the brought to you where it itself is activated overload is in general it's like AI yes so in general I talk about our team we refer to this distributed IJ in a decentralized fashion so essentially you have in a community of nose that come together and even though each one may not be trusted and otherwise certain thresholds of the set of nodes and he behaves properly then and the system can essentially achieve certain properties for example in the distributed I just I think you have you can maintain a mutable log and you can ensure that for some of the transactions actually I'll create a pound and then it's immutable and so on so first of all what's the ledger so it's a it's like a database it's like a data entry and so distributed ledger is something that's maintained across or is synchronized across multiple sources multiple nodes multiple notes yes and so where is this idea now how do you keep okay so it's important ledger a database to keep that to make sure so what are the kinds of security vulnerabilities that you're trying to protect against in the context of this the distributed ledger so in this case for example you don't want to some malicious nose to be able to change the transaction logs and in certain cases account double spending like your also calls you can also cause different views in different parts of the network and so on so the ledger has to represent if you're capturing like financial transactions has to represent the exact timing and the exact occurrence and no duplicates all that kind of stuff has to be represent what actually happened okay so what are your thoughts on the security and privacy of digital currency I can't tell you how many people write to me to interview various people in the digital currency space there seems to be a lot of excitement there and it seems to be some of it to me from an outsider's perspective seems like dark magic I don't know how secure I think the the foundation from my perspective of digital currencies that is you can't trust anyone so you have to create a really secure system so can you maybe speak about how well your thoughts in general about digital currency is and how you how it can possibly create financial transactions and financial stores of money in the digital space so you as security and privacy and so so again as I mentioned earlier in security we actually talk about two main properties and the integrity and confidentiality and so there's another one for availability you want the system to be available but here for the question you ask let's just focus on integrity and confidentiality yes so so for integrity of this distribution essentially as we discussed we want to ensure that's the different nose and right so they have this consistent video usually it's down through we call a consensus protocol and that's the establish share the view on this leche and that you cannot go back and change this immutable and so on so so in this case then the security often refers to this integrity property and essentially you're asking the question how much work how how can you attack the system so that the attacker can change the lock for example right how hard is it to make an attack like that yes right and then that very much depends on the the consensus mechanism the how the system is built and now that so there are different ways to build these decentralized systems and people may have heard about the term Scout like proof-of-work you prefer take you this different mechanisms and really depends on how how the system has been built and also how much resources how much work has gone into the network to actually say how secure it is so for example if you talk about like in the coins for what system is so much electricity it has been burnt so there's differences there's differences in the different mechanisms and the implementations of a distributed ledger used for digital currency also there's Bitcoin is a whatever there's so many of them and there's underlying different mechanisms and there's arguments I suppose about which is more effective which is more secure which is more what amount of resources needed to be able to attack the system like for example what percentage of the nose do you need to control our compromise in order to write to change the log and those are things do you do you have a sense if those are things that can be shown theoretically through the design of the mechanisms or does it have to be shown empirically by having a large number of users using the currency I see so in general for each consensus mechanism you can actually show theoretically what is needed to be able to attack the system of course there are there can be different types of attacks as weepy and discuss at the beginning and so that and it's difficult to gave like you know a complete estimate like really how much is needed to compromise the system but in general right so there are ways to say what percentage of the knows you need to compromise and so on so we talked about integrity so on the security side and then you also mentioned can the privacy or the confidentiality side does it have some of does it have some of the same problems and therefore some of the same solutions that you talked about and the machine learning side with differential privacy and so on yeah so actually in general on the public ledger in this public decentralized systems and actually nothing is private so all the transactions posters on the library anybody can see so in that sense there is no confidentiality and so usually all you can do is then there are the mechanisms that you can built in to enable confidentiality are privacy of the transactions and the data and so on that's also some of the work and that's both my group and also my startup and does as well what's the name you start o Asus labs Oasis labs and so the confidentiality aspect there is even though the transactions are public you want to keep some aspect confidential of the identity of the people involved in the transactions or what what is their hope to keep confidential in this context so in this case for example you want to your nipple like private confidential transactions even so so there are different and essentially types of data that you want to keep private are confidential and you can utilize different technologies including your knowledge proofs and also secure computing and techniques and to hide the right who is making the transactions to whom and the transaction amount and in our case also we can enable like confidential smart contracts and so that's you don't know the data and the execution of the smart contract and so on and we actually are combining these different technologies and to going back to the earlier discussion we had enabling like ownership of data and privacy of data and so on so so at Oasis labs we're actually building what we call a platform for responsible data economy to actually combine these different technologies together and to enable secure and privacy-preserving computation and also using the library to help provide immutable log of users ownership to their data and the policies they want the data to adhere to the usage of the data to adhere to and also how that it has been utilized so all this together can build we can a distributed secure computing fabric that helps to enable a more responsible data economy other things together yeah wow those eloquent okay you're involved in so much amazing work that we'll never be able to get to but I have to ask at least briefly about program synthesis which at least in a philosophical sense captures much of the dreams of what's possible in computer science and the artificial intelligence first let me ask what is program synthesis and can ural networks be used to learn programs from data so can this be learned some aspect of this synthesis can it be learned so program synthesis is about teaching computers to write code to program and I think it has one of our ultimate dreams or goals and you know I think Andreessen talked about software eating the world so I say once we teach computers to write software I had to write programs then I guess computers yeah exactly so yeah and also for me actually um when I you know shifted from security to more AI a machining program synthesis is program scenes in adversarial machining these are the two fields that I particularly focus on like program synthesis one of the first questions that I actually started what are seeking just as a question oh I guess with from the security side there's a you know you're looking for holes and programs so as at least see small connection but why what was your interest for program synthesis as because it's such a fascinating such a big such a hard problem in the general case why program synthesis so the reason for that is actually when I shifted my focus from security into AI machine learning and actually one of my main motivation at the time and is that even though I have been doing a lot of working security and privacy but I have always been fascinated about beauty intelligent machines and that was really my main motivation to spend more time in AI am a Shalini is as I really want to figure out how we can build intelligent machines and to help us towards that goal program synthesis is really one enough I would say the best domain to work on I actually call it's like programming synthesis it's like the perfect playground for building intelligent machines therefore artificial general intelligence yeah um well it's also in that sense not just a playground I guess it's it's the ultimate test of intelligence because yes I think I think you can generate so neural networks can learn good functions and they can help y'all in classification tasks but to be able to write programs right that's that's the epitome from the machine side that's the same as passing the Turing test and natural language but with programs it's able to express complicated ideas to reason through ideas and yeah and boil them down to algorithms yes exactly is that credible so can this be learned how far are we is there hope what are the open challenges questions and we're still at an early stage but already I think you we have seen a lot of progress I mean definitely we have you know existence proof just like the humans can write programs so there's no reason why computers cannot write programs and so I think that's definitely an achievable goal it's just how long it takes and then and even today we actually have you know the program synthesis community especially the program synthesis by learning our way College neural program synthesis community is still very small but the community has been growing and we have seen a lot of progress and in limited domains I think actually program synthesis is ripe for real-world applications so actually was kind of amazing I was at giving a talk it's also here it's a rework we worked you planning something actually so I give another talk at the previously rework conference in deep reinforcement learning and then I actually met someone from a startup and the CEO of the startup and when he saw my name he recognized and he actually said one of our papers actually had they have put the had actually become a key products and that was program synthesis in that particular case it was natural language translation translating natural language description into psycho Cory's oh wow that that direction okay right so yeah so you program since this is in limited domains in well specified domains actually already we can see really great great progress and applicability in the real roads so domains like as an example you said natural language being able to express something to just normal language and it converts it into a database sequel SQL query right and that's how how solve the problem is that because that seems like a really hard problem okay eliminate domains actually it can work pretty well and now this is also a very active domain after research at the time I think one he saw our paper at the time we were the state of the Arts yeah and that task and since then actually now there has been more work and with even more sophisticated assets and so but I I think I wouldn't be surprised that's more of this type of technology really getting to the real worlds that's exciting in the near term being able to learn in the space of programs is super exciting I still yeah I'm still skeptical because I think it's a really hard problem progress and also I think in terms of the your ass about open challenges I think the domain is full of challenges and in particular also we want to see how we should measure the progress in the space and I would say mainly three main I'll say metrics so one is a complexity of the program that we can synthesize and that will actually have clear measures and just look at you know the past publications and even like for example I was at the recent Europe's conference now there is actually very sizable like session dedicated to program since this is vicious or even neural progress today which is great and and we continue to see the increase like I think they were sizable it's five people and they will all win touring awards one day like it so we can see increase in the complexity of the program is that these synthesized sorry - is it the complexity of the actual text of the program or the running time complexity which complexity over how complexity after task to be synthesized and the complexes are after the actual synthesize the programs so you so the lines of code even for example okay I got you but it's not the theoretical upper bound of the running time of the day and you can see the complexity in decreasing already oh no meaning we want to be able to synthesize monomer complex programs bigger and bigger programs so we want to see that's we want to increase I have to think through because I thought of complexity is you want to be able to accomplish the same task with a simpler and simpler program no we are not doing that okay it's more it's more about how complex a task right we can see the exotic being able to synthesize programs learn them for more and more difficult right so for example initially our first working program synthesis synthesis was to translate natural language description into really simple programs called if TTT if this then that so given a trigger condition what is the action you should take so that program is a super simple you just Andy identify the trigger conditions and the action yeah and then later on with the secret queries that gets more complex and then also we started to synthesize programs with loops and know anything could synthesize recursion it's all over actually yeah 1fi works actually it's already rechristen you're complexity and the other one is generalization like one-way training I want to learn programming synthesizer in this case and neural programs to synthesize programs then you wanted to generalize so for a large number of inputs to be able to write generalize to previously and C inputs got it and so so someone for the work who waited earlier learning recursive new programs actually showed that recursion actually is important and to learn and if you have recursion then for certain and set of tasks we can actually show that you can actually have perfect generalization and so right so that one the best paper Awards that I clear earlier and so that's one example of we want to learn these you know programs that can generalize better but that works for a certain task with certain domains and there is question how we can essentially develop more techniques that can and have generalization for wider set of domains and so on so that's another area and then and then the the third challenge I think will it's not just for programming synthesis is also cutting across other fields in machine learning and also including like deep reinforcement and in particular is that this adaptation is that we want to be able to learn from the past and tasks and training and so on to be able to solve new tasks so for example in program synthesis today we still are working in the setting way given a particular task we change the right model and to solve this particular task but that's not how humans work like the whole point is we train a human than you can then program to south new tasks right exactly and just like we don't want to just change agent to play a particular game hey it's Atari ice ago whatever we want to train these agents that can and essentially extract knowledge from the past learning experience to be able to adapt to new new tasks and solve new tasks and I think this is particularly important for program synthesis yeah that's the whole point that's the whole dream of progress this is your learning a tool that can solve new problems right exactly and I think that's a particular main that as a community we need to put more emphasis on and I hope that we can make more progress today as well awesome I think there's a lot more to talk about but let me ask that you also had a very interesting and we talked about rich representations he had a rich life journey you did your bachelor's in China and your masters and PhD in the United States CMU and Berkeley are there interesting differences I told you I'm Russian I think there's a lot of interesting difference between Russia and the United States are there in your eyes interesting differences between the two cultures from the silly romantic notion of the spirit of the people to the more practical notion of how research is conducted that you find interesting or useful in your own work of having experienced both that's a good question I think so I I started in China for my undergraduate and that was more than 20 years ago there's been a long time is there echoes of that time I think even more so maybe something that's even be more different for my experience and a lot of computer science researchers and practitioners is that so for my undergraduate studies physics very nice and then I switch to a computer science in graduate school what happened was there was there is there another possible universe where you could have become a theoretical physicist at Caltech or something like that that's very possible some of my and undergrad classmates then the later studies physics account there 15 physics from these schools from yeah from tough physics programs so so you you switch to I mean in that from that experience to doing physics in your bachelor's how what means you decide to switch to computer science and computer science had arguably the best university one of the best universities in the world for computer science and with Carnegie Mellon especially for the grad school and and so on so what ii only 10 mighty just kidding okay I had Authority and know what what was the choice like and what was the move to the United States like what was that whole transition and if you remember if there's still echoes of some of the spirit of the people of China in you in New York it's like three questions so yes I guess okay the first transition from physics to computer science yes so when I first came to the United States I was actually in the physics ph.d program at Cornell yeah I was there for one year and then I switched to computer science and I was seeing the PC program at kind of give a loan and so okay so the reasons for switching so one thing so that's why I also mentions that about this difference in backgrounds about having studied physics yes first in my undergrad um actually really I really did enjoy my undergrads time and education in physics I think that actually really helped me in my future work in computer science actually even for machine learning a lot of machine learning stuff the the core machining methods many of the magic for honest most most of everything came from physics I was I think I was really attracted to physics and it was it's really beautiful and educated physics is the language of nature and I actually really remember like one moment in my undergrads like I did my undergrad in Chinua and I used to study in the library and I clearly remember like one day I was sitting in a library and I and I was like writing my notes and so on and I got so excited that I realized that if you just from a few simple axioms a few simple laws I can derive so much it's almost like I can't derive the rest of the world yeah there's the universe yes yes so that was like amazing do you think you have you ever seen or do you think you can rediscover that kind of power and beauty and computer science in the world that yes that's very interesting so that gets to you know the transition from physics to Versailles and it's a it's quite different for and for physics in in Cresco actually things changed so one is I started to realize that when I started doing research in physics at the time I was doing theoretical physics and a lot of its the you still have the beauty base very different so I have to actually do a lot of simulation so essentially I was actually writing in some in some cases writing a fortune Harold fortune yes to actually write do like do simulations and so on that was not not exact I I enjoy it's doing and also at the time from talking with the senior you know students in the program I realized many of the students actually were going off to work Wall Street and and so on and so and I've always been interested in computer science and actually essentially taught myself the C programming program right when in college and college somewhere for fun learning to do C programming you know in physics at the time I think now the programming profit has changed but at the time really the only class we had in in Hoosick amir science education was introduction to africa to computer science or computing and fortune 77 there's a lot of people that still use Fortran I'm actually if you're a programmer out there I'm looking for an expert to talk to about Fortran they seem to there's not many but there's still a lot of people to still use Fortran and still a lot of people these cobalt I realized instead of just doing programming for doing simulations and so on that I may as well just change to computer science and also one thing I really like and that's a key difference between the two as in computer science is so much easier to realize your ideas if you have idea you're writing it up you're cut it up and then you can see it's actually bring it to life quickly it's your life wasting physics if you how good theory you you have to wait for the experimentalist to do the experiments and to confirm the theory and things just take so much longer and and also the reason I in physics I decided to do theoretical physics it was because I had my experience with experimental physics first you have to fix the equipment fixing the equipment first so offensive equipment so there's a lot of it yeah he's have to collaborate with a lot of people takes a long time yes messy so I decided to switch to computer science and the one thing I think maybe people have realized is that for people who study physics actually it's very easy for physicists to change to do something else yes I think physics provides a really good training and yeah so actually it was very easy to switch to computer science but one thing going back to your earlier question so one thing I should you realize so there is a big difference between commune sense and physics away physics you can derive the the whole universe from just a few simple laws and computer science given that a lot of it is defined by humans the systems that you find by humans and and artificial I can essentially create a lot of these artifacts and so on and it's it's not quite the same you don't derive the computer systems with just a few simple laws you actually have to see there's historical reasons why our system is builds and designs one way versus the a day there's a lot more complexity or less elegant simplicity of e equals mc-squared that kind of reduces everything down to his beautiful fundamental equations but what about the move from China to the United States is there anything that still stays in you that's contributes to your work the fact that you grew up in another culture so yes I think especially back then it's very different from now so you know now they actually I see these students coming from China and even an aggressor actually they speak fluent English it was just you know like amazing and they have already understood so much of the culture in the US and so on and it was to you was all foreign it was it was a very different time at a time actually even we didn't even have access to email right not to mention about the wealth yeah I remember I had to go to you know specific like you know privileged several rooms too much knowledge about the Western world and actually at the time I didn't know actually the the in the US the West Coast weather is so much better than the yeah things like that actually it's very it's very yeah but now it's so different at the time I I would say there's also a bigger culture difference because there's so much less opportunity for shared information so it's such a different right I meant world let me ask me be a sensor question I'm not sure but I think you're not in similar positions is I've been here for already 20 years as well and looking at Russia from our perspective and you looking at China in some ways it's a very distant place because it's changed a lot but in some ways you still have echoes you have still have knowledge of that place the question is you know China is doing a lot of incredible work in AI do you see please tell me there's an optimistic picture you see where the United States and China can collaborate and sort of grow together in the development of AI towards you know there's different values in terms of the role of government and so on of ethical transparent secure systems we see it differently in the I States a little bit than China but we're still trying to work it out do you see the two countries being able to successfully collaborate and work in a healthy way without sort of fighting and making it an AI arms race kind of situation yeah I believe so and I think it's science there's no border and the advancement of technology helps everyone helps the whole world and so I certainly hope that the two countries will collaborate and I certainly believe so do you have any reason to believe so except being an optimist so first again like I said science has no borders and especially science doesn't know board borders right and you believe that will you know in this in the former Soviet Union during the Cold War yeah so this is the other point I was going to mention is that especially in academic research everything is public like we write papers we open source codes and others in the public domain it doesn't matter whether the person is in the u.s. in China or some other parts of the world and they can go on archive and look at the latest research and results so that openness gives you hope yes me too and that's also how as a world we make progress the best so apologize for the romanticized question but looking back what would you say was the most transformative moment in your life that maybe made you fall in love with computer science you said physics you remember there was a moment where you thought you could derive the entirety of the universe was there a moment that you really fell in love with the work you do now from security to machine learning to program synthesis so maybe as I mentioned actually in college a one summer I should tell myself programming see yes you just read a bug don't tell me you fell in love with computer science by programming and see remember I mentioned when one of the draws for me to come here sense is how easy it is to realize their ideas so once I you don't read the book started like it taught myself how to program and see immediately what what did I do like I programmed two games um ones just simple like it's a go game like it supports you can move the stones and so on and the other one actually programmed the game that's like a 3d Tetris it was a to not to be a super hard game to play it's obvious the standard 2d Tetris it's actually a 3d thing but I can realize wow you know I just had these ideas to try it out and then you can just do this so that's the one I realized wow this is amazing yeah you can create yourself from nothing to something that's actually out in the real world so let me ask let me ask a silly question or maybe the ultimate question what is to you the meaning of life what what gives your life meaning purpose fulfillment happiness joy okay these are two different questions very different yeah it's easy that you asked this question maybe this question is probably the question that has follows me and follow my life the most have you discovered anything and you satisfactory answer for yourself is there something is there something you've arrived at you know that there's a moment I've talked to a few people who have faced for example a cancer diagnosis or faced their own mortality and that seems to change their views and it it seems to be a catalyst for them removing most of the crap that the of seeing that most of what they've been doing is not that important and really reducing it into saying like here's is actually the few things that really give me give meaning mortality is a really powerful catalyst for that it seems like facing mortality whether it's your parents dying or somebody close to you dying or facing your own death for whatever reason or cancer and so on yeah in my own case I didn't need to face mortality and I think there are a couple things so one is like who should be defining the meaning of your life right is there some kind of even greater things than you who should define the meaning of your life so for example when people say that searching the meaning for our life is is there some there is some outside voice or is there something you know a set of you who actually tells you you know some people talk about oh you know this is what you have been born to do right right like this is your destiny um so who right so that's the one question like who gets to define the meaning of your life should you be finding some other thing some other factor to define this for you always something actually it's just entirely where you define yourself and it can be very arbitrary yeah so in inner and inner voice or an outer voice whether it's it could be spiritual religious - with God or some other components of the environment outside of you or just your own voice do you have up do you have an answer there and so you know you know the long period of time of thinking and searching even searching through outsides right you know voices are factors outside of me yeah so that I have and so I've come to the conclusion and realization that it's you yourself that you finds the meaning of life yeah that's a big burden no isn't it right so then you have the freedom to define it yes and and another question is like what does it really mean by the meaning of life right um and also whether the question even make sense absolutely and you said it somehow distinct from happiness so meaning is something much deeper than just any kind of emotional any any kind of contentment or joy whatever it might be much deeper and then you have to ask what is deeper than that what is what is there at all and then the question starts being silly right and also you can say it's deeper but you can also say it's a shallow depending on how people want to define the meaning of their life so for example most people don't even think about this question then the meaning of life to them it doesn't really matter that much and also whether knowing the meaning of life and whether actually helps y'all love to be present area or whether helps your life to be happier and these actually are often questions is not worse most questions open I tend to think that just asking the question as you mentioned as you've done for a long time is the only that there is no answer and asking the question is a really good exercise I mean I have this for me personally I've had the kind of feeling that creation is a like for me has been very fulfilling and it seems like my meaning has been to create and I'm not sure what that is like I I don't have a single lot of kids I would love to have kids but I also sounds creepy but I also see sort of he said see programs I see programs as little creations I see robots as little creations I think those are met those of those bring and then ideas theorems and and are creations and those somehow intrinsically like you said bring me joy I think they do to a lot of these scientists but I think they did a lot of people so that to me if I had to force the answer to that I would say creating new things yourself for you for me for me for me I don't know but like you said as he keeps changing is there some answer that some people they can I think they may say it's experience rights like their meaning of life all right they just want to experience to the richest and full as they can and a lot of people do take that path yes seeing life is actually a collection of moments and then trying to make the richest possible that's filled those moments with the richest possible experiences yeah right and for me I think it's certainly we do share a lot of similarity here like the creation is also really important for me even from you know the things that I've already talked about even like you know writing papers and these are our creations as well and I have not quite thought whether that has really the meaning of my life like in a sense also that maybe like what kind of things should you create there's so many different things that you could create and also you can say another view is maybe growth is it's related but different from experience growth is also maybe type of meaning of life it's just you try to grow every day try to be a better self every day and and also ultimately we are here it's part of the overall evolution the right the world is evolving it's funny it's funny that the growth seems to be the more important thing than the thing you're growing towards it's like it's not the goal it's the the journey to it sort of it's almost it's almost when you submit a paper it's there's a sort of depressing element to it not to submit a paper but when that whole project is over I mean there's a gratitude there's a celebration and so on but you're usually immediately looking for the next thing yeah the next step right it's not it's not that status that at the end of it is not the satisfaction is the the hardship the challenge you have to overcome the growth through the process it's something it's somehow probably deeply within us the same thing that drove that drives the evolutionary process is somehow within us with everything the way the way we see the world since you're thinking about this so you're still in search of an answer I mean yes and no in the sense that I think for people who really dedicate time to search for the answer to ask a question what is the meaning of life it does not as we bring your happiness yeah it's a question and we can say right like weather is a well-defined question and and on the other and but on the other hand given that you get two answers yourself you can define it yourself sure I can't just you know give it answer and in that sense yes it can help and like it's like we discussed if you say oh then my meaning of life is to create are to grow then then yes then I think they can help but how do you know that that is really the meaning of life are the meaning of your life it's like there's no way for you to really answer the question sure but something about that certainty is liberating so if it might be an illusion you know you might not really know you might be just convincing yourself falsely falsely but being sure that that's the meaning the there's something there's something liberating in that in that there's something freeing in knowing this is your purpose so you can fully give yourself to that without you know for a long time you know I thought like isn't it all right like why what's how do we even know what's good and what's evil like it isn't everything just relative like how do we know you know the the question of meaning is ultimately the question of why do anything why is anything good or bad why is anything moment then you start to I think just like you said I think it's a really useful question to ask but if you ask it for too long and too aggressively I mean not be so protect it not be productive and not just for traditionally society to find success but also for happiness it seems like asking the question about the meaning of life is like a trap is uh were destined to be asking we destined to look up to the stars and ask these big white questions we'll never be able to answer but we shouldn't get lost in them and that's probably the that's at least a lesson I picked up so far I'm noting that topic let me just add one more thing so it's interesting so actually so sometimes yes it can help you and to focus so when I when I shifted my focus more from security to a I am a Sunni at the time the actually one of the main reason why I did that was because at the time I thought my mini the meaning of my life and the purpose of my life is to build in hydrogen machines and that's and then your inner voice said that this is the right this is the right journey to take to build intelligent machines and that you actually fully realized you took a really legitimate big step to become one of the world class researchers to actually make it to actually go down that journey yeah that's profound that's profound I don't think there's a better way to end a conversation than talking for for a while about the meaning of life done it's a huge honor to talk to you thank you so much for talking today thank you thank you thanks for listening to this conversation with Dawn song and thank you to our presenting sponsor cash app please consider supporting the podcast by downloading cash app and using collects podcast if you enjoy the spot guest subscribe on YouTube review it with five stars on Apple podcast supported on patreon or simply connect with me on Twitter Alex Friedman and now let me leave you with some words about hacking from the great Steve Wozniak a lot of hacking is playing with other people you know getting them to do strange things thank you for listening and hope to see you next time you\n"