Talks on production machine learning from Toloka and Mesh AI
After a brief stint of online-only action during January, it was a relief to welcome members back to MLOps London in person again in March. With covid infections dropping across the UK and the world beginning to return to normal it was brilliant to see engineers, data scientists and AI enthusiasts all gathering under one roof again.
After the customary food, drinks and networking (the chicken & pesto baguette was a banger), the event kicked off with a talk from Magdalena Konkiewicz from Toloka.ai.
The focus of Magdalena’s talk was how machine learning needs to become more “data centric” and how practitioners need to spend less time tweaking the model and more time worrying about data quality. The topic, of course, has become very popular of late, but Magdalena also showed how crowdsourcing and labeling best practices can help you gather quality data in an efficient manner.
You can watch Magdalena’s full talk here and download her presentation here.
After a short break to collect more hop-based refreshments and discuss the previous talk, we dived into our second talk which was delivered by Sean and Mo from Mesh AI.
Mesh AI is a consultancy firm who deliver AI projects to businesses. The clients they work with, more often than not, undergo an MLOps transformation as part of that work and Sean and Mo were able to share their experiences of what it’s like to work on these projects. An important note was that cultural change is often harder to establish than the tooling or processes.
One highlight that stood out was a North Sea oil and gas client who took almost a full year to provide the data required for a project. When they did, the data consisted of over 65,000 files. That’s either a data scientist’s dream or worst nightmare, depending on how much they enjoy data exploration.
You can watch the full talk from Mesh AI here and download their presentation here.
Our next MLOps London event will take place on 10th May 2022 where we’ll be talking about feature stores and other production ML techniques. As previously, the event will be in-person in London but also streamed live online for those unable to travel. See you there!