The Browser, The Machine Learning and The Future we might not have imagined just less than a year ago
summary
Timeline of concrete progress from 2015 to 2017: JavaScript moves from hand‑built nets and convnetjs/brainjs experiments to numjs for matrices, GPU.js and WebDNN/WebGL transpilation, tensorfire and deeplearn.js running pretrained models as snapshot of weights in the browser. Benchmarks jump from 6.5× to 52× to 162× speedups. Demos include image captioning and real‑time style transfer via fragment shader implementations and WebGL shaders. Discussion flags black‑box limits and ethical risk when delegating decisions.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 3:40Machine learning curiosity
- 7:19Talk purpose and goals
- 9:50Science and ML magic
- 12:17How neural networks learn
- 14:27JavaScript ML struggles
- 16:55Build neural network in JS
- 18:02Higher-level ML libraries
- 20:30JS math libraries
- 21:33Neural network snapshots
- 22:35GPU acceleration in browsers
- 23:57GPU.js and WebGL
- 24:49WebDNN optimized models
- 25:50Machine Labs Python
- 26:58TensorFire demos
- 28:46DeepLearn.js creativity
- 29:55ML in JavaScript future
- 30:17AI ethics and responsibility
- 32:48Key takeaways
related talks
35mThe Future of JavaScript & Machine Learning
15mThe future of machine learning using JavaScript
32mDeep Learning in JS
48mDeep Learning in the Browser
41mEscaping the Cloud: High-Performance AI in your Browser
24mThe Future of Machine Learning & Javascript
15mMachine learning for next gen web apps with TensorFlow.js
29mSuperpowers for next gen web apps: Machine learning