Data science in the browser: DX & UX
summary
Shows a concrete pipeline: running TensorFlow.js t-SNE and UMAP on MNIST and clothes images inside the browser, moving heavy work into web workers and offscreen canvas so GPU compute happens off the main thread, and preserving results with localStorage caching to avoid recompute. Packs that into React component wrappers (open source as React Flatland), reports UMAP converges faster and reveals sensible clusters, notes image-loading and browser support limits but demos mostly run and feel interactive.
AI-written summary · about
chapters (13)
- 0:00Start
- 0:28Data Science in the Browser
- 3:21Reusable React Components
- 6:21Mnist Hand Drawn Digits Data Set
- 8:03Componentizing
- 10:03Web Workers
- 11:54React
- 12:19How To Bundle Workers
- 13:03Offscreen Canvas
- 13:56Caching
- 15:43Demos
- 21:56Linear Optimization
- 28:52Have You Used Web Workforce To Solve Other Problems
related talks
18mEscaping flatland: a romance of data science in the browser
42mA Front-end dev's hobbyist approach into Machine Learning
34mMachine Learning in the browser
37mVisualizing Large Datasets with JavaScript Using deck.gl
14mFuture-proof web apps with Angular and TensorFlow.js
28mLightning Talks - Day 2
23mNode.js = Machine Learning
37mPainting with Data