Braden Hancock is a machine learning researcher, developer, and entrepreneur. Before co-founding Snorkel AI in 2019, he spent four years developing new programmatic approaches to labeling, augmenting, and structuring training data with the Stanford AI Lab, Facebook, and Google.
In this session, co-founder of Snorkel AI, Braden Hancock, shares how ML can be made practical with open source!
Introduction and outline - 0:00
Impractical machine learning assumptions (Braden explains the differences between ML in academia and ML in industry. Three impractical academia assumptions: a large high-quality task-specific training dataset, an infinite pool of qualified annotators, a static test distribution.) - 1:02
Snorkel AI: how it works? (Snorkel OSS was created to make ML practical again. Braden takes us from the beginnings at Stanford AI Lab and explains what Snorkel does differently.
Goals: make labeling, building and managing training data programmatic instead of manual. Combine weak, noisy labeling functions in an unsupervised way and make them strong. Be able to generalize to new examples not covered by labeling functions.) - 4:36
Snorkel AI: does it work? (How Snorkel does on various metrics: speed, cost, privacy. Industry adoption. Academic leaderboards. Case studies.) - 9:35
What did we learn? The four “I”s. Interfaces: user experience can be improved with a GUI, common labeling function types can be templatized. Infrastructure: enterprise-level support is needed for important software. Interactions: creating interaction points for domain experts, data scientists and developers. Intuition: a new interface for ML must come with new best practices, tips and tricks.) - 13:12
Snorkel Flow - a platform for building AI applications. (How the four “I”s moved Snorkel to build Snorkel Flow. Braden presents 4 guiding principles behind Snorkel Flow. ) - 15:56
Closing remarks and how to connect. - 19:32
Share your questions and comments below!