Accelerating autonomous vehicle solutions with SageMaker, featuring Lyft
summary
Lyft Level 5 migrates heavy ML training from self-managed Kubernetes to Amazon SageMaker and cuts turnaround from days to minutes/hours by using distributed training (PyTorch DistributedDataParallel, up to dozens of nodes) and a small CLI that builds/pushes containers to ECR. Data labeling uses SageMaker Ground Truth for 2D images and 3D point clouds with auto-labeling that claims up to 70% cost reduction. Using managed spot instances and SageMaker’s one‑click APIs, Lyft trains three times more models while reducing total cost.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:52Challenges in AV ML
- 3:59SageMaker capabilities overview
- 9:20Lyft Level 5 overview
- 16:12Lyft’s ML workflow
- 19:44Switching to SageMaker
- 23:18Results and cost savings
- 26:10Technical conclusions
- 26:33SageMaker demo
- 29:44Lessons learned
related talks
19mDrive transformation through machine learning
33mBuild, train, and deploy ML models using Amazon SageMaker
39mTrain ML models at scale with Amazon SageMaker Training
19mUsing Amazon SageMaker to develop and deploy ML models
37mStreamlining Machine Learning Operations
37mStreamlining Machine Learning Operations
31mUse AutoML to create high-quality models with just a few clicks
28mFuture of Data Platforms: making use-case driven...