MNIST Classifier

This interactive demo implements a vanilla feedforward neural network to solve one of the most famous machine-learning tasks: classifying the MNIST dataset . The classifier is implemented from scratch in pure JavaScript; all training and inference is performed locally in your browser, simply train the model to have it recognize a handwritten digit.

Sandbox

0123456789
-
[untrained network]

Overview

Initializing…
784
30
10

Layers configuration

  • Hidden layer 1

Training configuration

Epochs
Mini-batch size
Learning rate

Statistics