Applied machine learning at Facebook
summary
Describes Facebook’s end to end ML stack with concrete builds and open releases: hardware servers for each pipeline stage including two storage platforms (flash and disk), CPU and GPU training platforms, and two CPU inference flavors (dual‑socket heavy hitter and single‑socket). Software is FB Learner with a FB Learner feature store, FB Learner Flow for scheduling and model selection, and epi learner predictor for deployment and regression tracking. Notes the merge of research and production frameworks into PyTorch 1.0.
AI-written summary · about
related talks
26mEwa Dominowska, Engineering Manager, Facebook @ MLconf SEA
06mOpening Address
50mDemocratizing AI - Back-fitting end-to-end machine learning
54mMachine Learning explained with Examples
19mRushin Shah, Engineering Manager, Facebook
42mDeep learning beyond the learning
35mA Course on Machine Learning for Software Developers
50mLearning Machine Learning