MNIST Digit Recognition

The Burn project enables building Pytorch style Deep Learning models. The following example serves predictions from a trained MNIST model in real time. Source code available from the Burn project (here).

Draw a digit (0-9) and watch a neural network classify it in real-time

Draw here

Prediction

Draw a digit to see prediction
Confidence Scores
0
0.0%
1
0.0%
2
0.0%
3
0.0%
4
0.0%
5
0.0%
6
0.0%
7
0.0%
8
0.0%
9
0.0%

About this demo

This demo runs a convolutional neural network entirely in your browser using WebAssembly. The model was built with Burn, a Rust deep learning framework.

  • Model: 2-layer CNN with ~260k parameters
  • Accuracy: 95.83% on MNIST test set
  • Inference: ~10ms on modern hardware
  • No data leaves your browser

Model will load on first classification (2.9 MB)