Machine Learning in the Browser
summary
Builds a browser-side neural net with tensorflow.js (formerly deeplearn.js), generates 10,000 data points of random RGB inputs and complementary RGB targets, and trains with mean squared error and gradient descent. Uses a small network, batch size 25, runs ~1,000 batches, and shows the loss shrinking and predictions becoming nearly indistinguishable from the target. Argues client-side benefits — latency, offline availability, and privacy — and provides code and a demo that works in the browser.
AI-written summary · about
chapters (14)
- 0:00Intro
- 1:05LIAN LI
- 4:35MACHINE LEARNING
- 6:43CLIENT-SIDE ML
- 12:41NEURAL NETWORK
- 13:48NODE
- 16:06TENSOR
- 16:56PLACEHOLDER
- 18:06FULLY CONNECTED
- 20:56CHECKLIST
- 22:26TRAINING
- 26:43OPTIMIZER
- 30:32DATA
- 34:59LINKS
related talks
35mMachine Learning in the Browser with deeplearnJS
34mMachine Learning in the browser
26mMachine Learning in the Browser with Deep Neural Networks
48mDeep Learning in the Browser
29mMachine Learning with Deep Neural Networks in the Browser
35mThe Future of JavaScript & Machine Learning
32mDeep Learning in JS
41mPractical Machine Learning for Front-End Developers