Machine Learning Platforms
summary
Argues that getting ML into production is about architecture not just algorithms and shows real tool chains: LLVM for retargeting, TensorFlow and PyTorch for training, and TensorFlow.js in browser for deployment. Demonstrates a MobileNet demo running locally with 3.3s → 49ms → 5ms → 3ms across CPU, WebGL, GTX1080 and Node, and a working WebRTC + camera inference pipeline for real‑time classification. Recommends edge/CDN placement and hosted APIs when training or latency constraints make sense.
AI-written summary · about
chapters (12)
- 0:00Introduction and welcome
- 1:03What is a machine learning platform
- 8:32Hardware's role in ML platforms
- 13:28Deployment architectures for ML
- 18:46Learning platforms for ML
- 23:23Productionizing ML models
- 30:10Java-based ML frameworks
- 35:10TensorFlow and TFX overview
- 38:07Caris and model export options
- 43:05TensorFlow.js and browser ML
- 47:39WebAssembly backend for ML
- 50:09Resources and closing remarks
related talks
15mThe future of machine learning using JavaScript
41mTensorFlow.js 101: ML in the Browser and Beyond
34mA Powerful Introduction to Machine Learning with Js
34mBuilding Machine Learning-powered Applications with JS
24mThe Future of Machine Learning & Javascript
23mThe new era of ML in JavaScript
20mTensorFlow.js Bringing Machine Learning to the Web and Beyond
06mOpening Address