TensorFlow.js Bringing Machine Learning to the Web and Beyond
summary
A library called TensorFlow.js runs training and inference in the browser and Node using WebGL fragment shaders for GPU acceleration, with converters for Keras/TensorFlow models and weights hosted on GCP and NPM. Pretrained models like pose net and bodyPix run locally (person segmentation, portrait blur), MobileNet hits ~23 ms per image (~30 fps) in Electron, Jetson Nano reports ~76 ms, and the headless node GL stack ships as a ~5–10 MB package versus CUDA builds ~250 MB.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:59TensorFlow Playground
- 2:09Privacy
- 2:27TensorFlowjs
- 2:55Designing the Library
- 4:59The Technical Stack
- 6:21Models
- 9:13Web Examples
- 11:13WeChat
- 11:32JavaScript
- 12:43NodeYes
- 13:44Demo
- 15:00Jetson Nano
- 15:34TF GS Biz
- 16:36Future of TensorFlow
- 18:45Data Library
- 19:32Thank you
related talks
48mTensorFlow.js
24mTensorFlow.js + Node
41mTensorFlow.js 101: ML in the Browser and Beyond
37mTensorFlow.js - Bringing ML and Linear Algebra to Node.js
31mIntroduction to Machine Learning in Tensorflow.js
15mMachine learning for next gen web apps with TensorFlow.js
31mDeep Learning at the Edge with TensorFlow.js
28mMachine Learning on the web with TensorFlow.js