Lightning talk: Machine Learning Intuitions for DeepLearn.js
summary
Concrete demo of DeepLearn.js on MNIST shows a convolutional network with matching layer dimensions and a final 10-way output, trained over ~300 epochs in ~76–77 seconds. The live model predicts a sample as digit 5 with ~97% confidence while confusing 3 and 8 at ~2% each. The talk explains inputs as grayscale matrices, training as label-driven error reduction, and notes that why deep nets work remains open research.
AI-written summary · about
related talks
44mMachine Learning Exposed
26mMachine Learning in the Browser with Deep Neural Networks
35mMachine Learning in the Browser with deeplearnJS
48mDeep Learning in the Browser
36mMachine Learning in the Browser
54mIntroduction to Understanding Deep Learning
35mThe Future of JavaScript & Machine Learning
27mDemystifying Deep Neural Networks