Inside NVIDIA’s AI Infrastructure for Self-Driving Cars

Explore NVIDIA's Project MagLev, an end-to-end AI platform for developing self-driving car software, in this 26-minute USENIX conference talk. Dive into the infrastructure supporting continuous data ingest from multiple vehicles producing terabytes of data hourly. Learn how autonomous AI designers iterate training of new neural network designs across thousands of GPU systems and validate their behavior over multi-petabyte-scale datasets. Discover the overall architecture for data center deployment, AI pipeline automation, large-scale AI dataset management, training, and testing. Gain insights into the development workflow, data lake overview, collection methods, active and targeted learning, ground truth labeling, traceability, and data preparation techniques used in NVIDIA's DRIVE software ecosystem.

Intro
The problem of selfdriving cars
NVIDIAs principles
Development workflow
Platform overview
Data lake overview
Collection
Active Learning
Targeted Learning
Ground Truth
Labelling
Examples
Labeling
Traceability
Export
Data Preparation