How to make ML teams 80% more productive.
summary
Claims an 80% productivity gain by automating an end‑to‑end continuous learning loop: petabyte‑scale data processed with Terra (Apache Beam + Google Dataflow) to build labeled training sets, automated air‑mining, simulation and a regression test suite that accumulates thousands of problematic scenes, training jobs that can run seven days, and city rollouts reused without restart. Platform features include lineage tracking, frequent checkpoints and caching for warm restarts, fault‑tolerant retries, and a Bazel‑driven local→cluster submission flow so experiments scale quickly.
AI-written summary · about
chapters (11)
- 0:00Introduction and background
- 1:57Building ML platform at Cruise
- 4:24Data access challenges and solution
- 7:28Continuous learning machine concept
- 10:38End-to-end automation workflow
- 14:29Lineage tracking and compliance
- 17:00Fault tolerance in pipelines
- 20:10Observability for debugging
- 21:35Local development and iteration
- 24:27Why continuous ML matters
- 25:32Open source ML platform
related talks
25mBuilding Scalable MLOps Pipelines That Actually Work
13mMLOps: From AI Research to Production-Ready Systems
31mProductionizing ML with ML Ops and Cloud AI
19mSustaining ML in the enterprise
38mFrom Experimentation to Products: The Production Machine Learning Journey
38mDistributed Processing for Machine Learning Production Pipelines
09mMachine Learning: Catalyst for Cloud-Native Digital Transformation
32mMaking Models Work: How an optimized ML lifecycle can drive better ROI