The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Live from TWIMLcon! Operationalizing ML at Scale with Hussein Mehanna - #306
Oct 8, 2019
Join Hussein Mehanna, Head of ML and AI at Cruise, as he shares his journey from Facebook to Cruise, detailing the evolution of machine learning platforms. He dives into the challenges of scaling ML with innovation, showcasing his insights on optimizing workflows and collaboration in tech giants. Hussein also discusses the complexities of autonomous vehicles and predicts the future of machine learning in enterprises, underscoring the role of cloud services and collaborative tools like Kubeflow.
AI Snips
Chapters
Transcript
Episode notes
Facebook ML Growth
- Hussein Mehanna joined Facebook in 2012 when only 20 people used ML.
- By the time he left, thousands of engineers used ML, training millions of models monthly.
Early Platform Efforts
- Initially, there wasn't a Facebook-wide ML platform, only team-specific tools.
- Mehanna focused on the platform and reduced model training time from 3 days to 21 hours, then to 2 hours.
FB Learner Evolution
- FB Learner evolved through three phases: Ad Learner, FB Learner, and FB Learner Flow.
- FB Learner Flow introduced workflows and a compositional approach that enabled scaling and extensibility.