Image Analysis and Model Training @ Scale on GKE

Seasoned ML developer, Sergei Makar-Limanov Ph.D. has built ML implementations at TripAdvisor, LevelUp, CNet, and most recently mabl.

Sergei will present how to build an online image analysis and training pipeline at scale on GCP, using an entirely serverless architecture. He'll discuss the use of Google Kubernetes Engine to cost effectively train and retrain models at scale over millions of images using GCP features like preemptible VMs. The challenges and benefits of combining the training pipeline into a real time, serverless training/prediction system with Cloud Dataflow, Pub/Sub, Functions, MemoryStore, and BigQuery will also be detailed.

Hosted at the Broad Institute

5:30 Doors open
6:00 Food and announcements (Boloco burritos and chips!)
6:30 Talk by Sergei Makar-Limanov Ph.D.
7:15 Q&A Session

Notes:
1. Parts of this session will be recorded. This is experimental, so we can't promise that a recording of the entire event will be available. It is something we are moving toward for future sessions.
2. The food selection will include a number of vegetarian burritos, but supplies will be limited to a first come, first served basis