The Event Experience: A Behind-the-Scenes Look at TuMoCon 2023
As I looked around the crowded room, I couldn't help but feel a sense of excitement and gratitude. The TuMoCon 2023 event was in full swing, with attendees from all walks of life gathered to discuss the latest developments in machine learning. The atmosphere was electric, with conversations flowing freely among the crowd.
I noticed a shrug being thrown around by someone in the audience, and I couldn't help but chuckle. It's moments like these that remind us that even in the tech world, we're still figuring things out. But as I took a glance at my notes, I saw that the app was still open for another month or so, providing ample time for attendees to dive into the latest resources and tools.
One of the most popular topics of discussion was differential privacy and federation around trying to use local devices for inference. It's clear that this is an area of great interest among machine learning professionals, with many hoping to preserve their data's privacy by keeping it on-device. As I listened to the conversation, I couldn't help but feel a sense of hope for the future of this field.
Meanwhile, another session was underway, discussing ways to make ML teams more efficient. The speakers dove into topics such as power-law distributions of team sizes and how that affects group dynamics. It's clear that there's still much to learn about how to bring together diverse groups effectively, but the enthusiasm in the room was palpable.
As I made my way through the crowd, I stumbled upon a conversation with an attendee who shared their experience with ML teams. They spoke of having "full-stack" engineers – unicorns who can handle both the technical and business sides of projects. It's clear that these individuals are in high demand, but also come with a price tag.
The discussion turned to the topic of agile methodologies for machine learning, and how they can be effective even when working on complex data sets. One speaker noted that units of work – stories, tasks, and such – should deserve to be first-class citizens in ML projects. It's an interesting perspective that challenges some of the conventional wisdom around data science.
As I continued to navigate the event space, I couldn't help but feel a sense of appreciation for the organizers' efforts. The little podlets they'd set up upstairs were filled with engaging conversations and thoughtful discussions. If you're interested in jumping into one of these podcasts, feel free to swing by and share your thoughts on the event or what you're working on.
The organizers also extended a huge thank you to their team members – my partner in crime, my wife Khalila, who put up with me during those early weeks of podcasting. The event management team, Carol and Linda, were instrumental in making this event happen. And last but certainly not least, the AV team, led by Amari and Mia, worked tirelessly behind the scenes to bring everything together.
Finally, a special mention goes out to the photographers who captured our moments in time. It's their work that allowed us to relive some of these memories long after the event has concluded. And so, I'd like to extend my gratitude once more to everyone involved – from the speakers to the attendees, and to every single one of you who made TuMoCon 2023 such a success.
It may be too early to say for certain, but it's clear that this was an event for the ages. One that will be remembered for years to come as a testament to the power of community and collaboration in driving innovation forward.
"WEBVTTKind: captionsLanguage: enall right everyone file on in they can't hear me outside it's only the folks that are in here that can hear me i get it what was the last unconference session that was in here data management nice important topic maybe it's all good no i'm i'm just uh closing things closing remarks here but the groups that are in the hinterlands are making their way over so we've got a couple minutes before we get started talk about twiml all right twiml ama what would you like to know about twiml stuff the whole point of twiml was to make it so that you hear what i hear global domination yes actually so two days ago was international podcast day who knew that anybody uh and maybe two days before that we realized the happenstance that we are on the all-time best seller tech podcast list on itunes this is pretty awesome thank you so actually uh so the summer before last i had my daughter who was uh now a sophomore uh but was a freshman going into her freshman year doing computer science do an experiment for us looking at um using cloud transcription services to transcribe the podcast we learned a lot in that experiment so our initial our plan going into this was she was just gonna take a bunch of services we had some transcribe file some some pot some uh audio and transcriptions and she was going to run these through a bunch of different services and kind of measure the results we learned that first of all the default settings for all of these services are basically useless right they have all different kind of modes you can put them in different channels and there's a voice and a video and stuff like that and you really have to spend a lot of time with each service trying to understand what its best settings are for your use case and then they all really struggled with technical conversations and accents and multi-party conversations uh it was not easy we don't have transc transcription up on the the podcast yet we do it uh from time to time but usually if we do if we're doing it uh for reference like for internal reference we'll do machine transcription and just kind of overlook its foibles uh but for uh things that matter where we're gonna publish it we always use a service that does it by hand so i'm sure someone will fix this problem and there are probably things that we haven't found yet but uh it wasn't quite ready for us uh how are we doing on time is carol in here are we good to go yep okay cool so um i thought maybe we'd start with a little bit of unconference wrap up i i wonder if everyone from the if that's in this room was just in this last unconference session and everyone else took the opportunity to escape but do we have some volunteers just to kind of share you know how many of you have been to unconferences before or done unconferences before so mostly first timers how was it it was a pretty cool addition to the agenda any uh takeaways from any of your sessions any volunteers we've got mic rudders we've got to use them hey everyone i'm robert uh i wasn't the organizer of the the data session just now but the organizer didn't turn up um so a few of ours which is okay it's an unconference yeah they voted with their feet it was probably better yeah um uh so basically uh no one has solved version in um unfortunately uh we didn't solve it in the unconference um but what about the people that claim that they've solved versioning um i mean there are vendors that have claimed they solved version in we'll we'll see how vendors are people too but um vendor companies are people too um but the the most common pattern that people agreed was working for for a lot of them was simply to use github uh to version uh the metadata uh around their uh their data sets um and then take care of data sets on the file systems themselves um so i guess the the big takeaway there would be if that's what you're doing right now um there's no consensus out there that there's anything better so you can feel good about it cool how about you hand the mic to my c and you can talk about your winning session yes um so open what was the session again ml infra it was built by versus open source um i have notes that i will be writing up and posting on my train ride back so in the next hour or two but it was a cool session people were discussing some good things about open source like a lot of community support particularly in stack overflow issues related to lack of maturity in open source ml infrastructure tools and the big takeaway from there for people looking to buy is know what you want before you go into a vendor conversation because then you're gonna maybe get sidetracked by what the vendor is offering that might not be a good fit either ways so cool i'll mention on that note that the app uh what what do you think about the app how did you like using the hoover app it's kind of top of the heap for conference apps right not too shabby yeah i saw a shrug back there i saw you well the app is still it stays open for at least another month or so so for example the blog post you're writing up you can post it in that conversation and folks can find it there uh also the slides uh that we've collected from presenters not all of the presenters have given us their slides um but we'll be chasing them down and all that and the ones that we have will be uploaded by monday and you just go into the session and the slides will be in there in the app uh any other unconference uh cool yeah so we had a really good group of people at the edge uh one and uh it's pretty early yeah so inference at the edge is really not that mature and we're still all kind of groping our way around but it really holds great potential so the real benefit of inference at the edges you can preserve privacy basically if the data never leaves your device then essentially you're guaranteed that that data won't be misused so we were basically discussing differential privacy and federation around trying to essentially actualize using uh the local device for inference so it's pretty good comfort uh pretty good uh pretty good group and that's very positive awesome john i could do the making ml teams more efficient session okay um we had a nice power law distribution of people team sizes ranging from one to two group uh experiences up to 35 and 100 so we that was really nice and we were small enough to fit around a little table in the pub that was also super nice it's really hard to hear people in these big open spaces i found um and so that we had a good session we um some of the take homes were that um full stack ml engineers are unicorns and uh even if you have one and he's a jerk you should uh hang on to them whoa that's very specific i know and there is a version of agile that works with ml and and you should embrace it um that was good news for me um oh it's it's just agile but for example uh the units of work in in machine learning uh deserve to be first class citizen you know stories and such and so forth um which a lot of data scientists don't like but it really helps your team anyone from that session remember anything else cool i guess that covers it cool so probably the most popular question that i've received today is when's the next turmoil con what do you all think about the conference nice awesome cool uh so we will keep you posted as to when the next fumikon is but uh in my opening remarks i showed that uh this event grew out of in no small part uh impromptu stuff that i've been doing at nurps for the past couple of years and for any of you who are also 10 academic from time to time i will be at nurps again this year and we'll probably do something a little bit bigger and more formal so keep an eye out for that the little podlets we've been doing have been awesome if you would like to jump on a few minute podcast and uh chat with me i'd love to talk to you and learn a little bit about what you're doing and you know we're posting those up we've set up upstairs right uh by the top of the stairs so if you are not running for caltrain and you can stick around for a few minutes and share your thoughts on the event or what you're up to i would love to talk to you um but other than that i really just want to say thanks to all of you for coming and uh thanks to the great team that helped put this event together uh first and foremost my partner in crime my wife khalila has been an amazing an amazing support for me in all things period but especially when i started this podcast three years ago and uh the first few weeks of it were like oh yeah i'm just gonna spend all of our weekends making this podcast and uh she put up with that for a few weeks which let it get started carol and linda are our event management team could not have done this without them i spent most of my time at this event walking around wondering when the other shoe was going to drop and no shoes ever dropped this was such a smooth event hopefully it was a smooth event for you i know the food was great the trains were running mostly on time for the most part and i really enjoyed it so thank you both for your help av if you had a chance to meet the av team was a great support as well photographers thank everyone thank you all um and uh amari and mia are outside kind of getting the the podcast booth ready um so they're not here to hear your applause but uh they are amazing as well so thank you all for coming to tuemocon and being part of the first event thank you\n"
