MNIST Neural Network Visualizer

Demo: https://mnist-nn-visualizer.vercel.app

• Model training
• Model inference on python
• Phase 1: Model inference Cloud Function
  • React client to draw digits
    • Mouse pointer events are not publishing consistently, adding (Bresenham's line algorithm)[https://www.wikiwand.com/en/articles/Bresenham%27s_line_algorithm] to fill in the gaps
    • Consider using html canvas instead of divs
  • Send image to endpoint (with debounce)
    • Setup REST API endpoint
    • Read weights from json file
    • Rebuild network structure
      • Layers
      • ReLU
      • Softmax
    • Load weights into network
    • Run inference
    • Return prediction and Tensor values to React client
  • Display prediction
• Phase 2: Client side inference
  • Setup numjs
  • Load weights
  • Run inference directly in browser
  • Display prediction
  • Display Tensor values

Resources

https://nextjournal.com/gkoehler/pytorch-mnist https://stackoverflow.com/questions/26432492/chessboard-html5-only https://www.wikiwand.com/en/articles/Bresenham%27s_line_algorithm https://github.com/lionelmessi6410/Neural-Networks-from-Scratch https://sim51.github.io/react-sigma/

License

GNU General Public License v3.0