FBLearner Flow is Facebook’s machine learning platform, used by over a dozen products within Facebook including Search, Ads and News Feed to train models aiming to deliver more relevant content to users. Tens of thousands of models are trained every week, using trillions of training examples. The system spans several areas, including ML infrastructure, algorithms, and applications built on top of the platform.
FBLearner Flow simplifies the process of building machine learning pipelines. Engineers use Flow back-end APIs to describe their workflows as directed acyclic graphs (DAGs) with typed inputs and outputs. The back end will take care of distributing the DAG, managing dependencies, failures, and shuttling intermediate outputs between substeps. The Flow front end provides a web UI that facilitates input verification, output visualization, parameter sweeps, and other operations.
In this session, we will demo FBLearner Flow and describe the system design principles and architecture.
Hussein Mehanna is an engineering manager at Facebook, where he founded and manages the Applied Machine Learning platform team. Hussein started as the original developer on the team, which quickly developed from an ads-focused ML platform to a Facebook-wide platform. Prior to Facebook, Hussein worked as a software engineer for Bing, Microsoft. He is a holder of a masters degree in speech recognition from the University of Cambridge, UK.
©2015, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • firstname.lastname@example.org
Apache Hadoop, Hadoop, Apache Spark, Spark, and Apache are either registered trademarks or trademarks of the Apache Software Foundation in the United States and/or other countries, and are used with permission. The Apache Software Foundation has no affiliation with and does not endorse, or review the materials provided at this event, which is managed by O'Reilly Media and/or Cloudera.