Machine Learning in the Browser with deeplearnJS
summary
Live in-browser demo generates arbitrary-size training data of normalized 3-element RGB arrays, builds a sequential dense network with 3 inputs and 3 outputs and three hidden layers, uses ReLU activation on the output, compiles with SGD + MSE and trains with configurable batch size and epochs; training runs on the GPU via WebGL-accelerated training (deeplearn.js / TensorFlow.js), and the model quickly approximates complementary colors, the demo working surprisingly well.
AI-written summary · about
chapters (25)
- 0:00Intro
- 2:50Follow me
- 3:13About me
- 4:03About machine learning
- 4:43Machine learning vs expert systems
- 6:17Experience vs repetition
- 6:59Advantages
- 8:18Problem
- 9:23Terminology
- 13:02Code
- 14:59Checklist
- 15:28Tensorflow Jazz
- 16:27Neural Networks
- 19:00Weights
- 20:23Layers
- 24:33Training
- 24:45Three Principles
- 25:11CostLoss Function
- 26:21Optimizer
- 27:36Flow Jet
- 28:52Running the algorithm
- 30:15Reducing the error
- 30:43Retraining during runtime
- 31:39Live demo
- 34:05Comedy Company
related talks
36mMachine Learning in the Browser
48mDeep Learning in the Browser
34mMachine Learning in the browser
29mMachine Learning with Deep Neural Networks in the Browser
32mDeep Learning in JS
26mMachine Learning in the Browser with Deep Neural Networks
35mThe Future of JavaScript & Machine Learning
39mRun Deep Learning models in the browser with JavaScript and ConvNetJS