Peer Beneath the Surface of the Earth
summary
Claims a new radar collects 50,000 points per second (not three) and Widmo builds a pipeline that turns one-row profiles into more than one million generated profiles for ML. Infrastructure uses CDK, Lambdas, Docker and SageMaker; training runs 100 random-config models then picks the best three and retrains. A custom cluster moves from single-region 8 V100s to distributed 8 regions to get 24 GPUs in a day. Multi‑cloud work with BigQuery/BigLake and saved $20,000 per quarter.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:09Unknown layers beneath cities
- 3:39Radar technology for mapping
- 4:46Custom GPU simulation cluster
- 6:12Infrastructure as Code with AWS CDK
- 7:58Generative AI for automation
- 14:36Machine learning pipeline
- 17:55Model deployment
- 20:52Multi-cloud with BigQuery Omni
- 25:17Data products and integration
- 26:46BigQuery Omni setup demo
- 32:03Data processing optimization
- 36:22Security best practices
- 39:11Security monitoring
- 40:15Q&A
related talks
44mScaling ML from 0 to millions of users
29mRunning Massively Parallel Deep-learning Inference Pipelines on Kub... Suneeta Mall & Martin Abeleda
48mAccelerate science, research, and innovation with AWS
28mScaling Machine Learning Training and Inference
20mHow the cloud changed the way I do analytics and research
44mBest practices for analytics and generative AI on AWS(ANT302)
29mMachine learning in production
30mMaster large-scale ML batch inferencing