Vanilla JS MNIST Neural Network Visualizer

Plain HTML, CSS and JavaScript. No external libraries.

1. Dataset

Put MNIST IDX files into data/. Then load them from Apache.

No dataset loaded.

2. Model

No model yet.

3. One sample

4. Training

Not training.

Network

Input layer is shown as the MNIST image. Only a few neurons per layer are drawn to keep it readable.

Prediction: -

Input image

Label: -

Selected value

Click a neuron or connection.

Training loss

Expected project structure

ann-vanilla-mnist-app/
  index.html
  css/styles.css
  js/*.js
  data/
    train-images.idx3-ubyte
    train-labels.idx1-ubyte
    t10k-images.idx3-ubyte
    t10k-labels.idx1-ubyte

The app also tries common gzip names such as train-images-idx3-ubyte.gz. Gzip loading needs browser support for DecompressionStream.