ML Infrastructure for Autonomous Vehicles
summary
Cruise runs over a petabyte of high‑resolution temporal sensor data, collects tens of gigabytes per mile, and expects hundreds of thousands of vehicles this decade; the ML platform automates long‑tail discovery and retraining via a continuous learning machine that mines faults, labels examples, trains models and safely validates deployments. A unified pipeline uses semantic steps and a Pythonic data source IR so the same graph powers online training and offline materialization, with distributed index building that drops a 20‑minute job to a few seconds using an array scheduling DSL.
AI-written summary · about
chapters (5)
- 0:00Start
- 2:56Planning overview: Deciding what to do
- 4:49The Long Tail Challenge
- 8:32Challenge - End to End Al Workflow
- 11:28One Platform Architecture
related talks
28mPredicting the Unpredictable Cruise’s Continuous Learning Machine
29mScaling ML Pipelines for Autonomous Driving with Ray
29mUsing Big Data And Ai In Transport And Logistics
44mML in Production
34mOperationalizing ML at Scale
27mHow to make ML teams 80% more productive.
32mThe ML system endeavors
31mDeep learning for self-driving vehicle technology