Neural network · MNIST
A small network learning to read handwritten digits from 0 to 9. Click any neuron to see the math behind it.
Across all 500 held-out images. Loss measures how much probability the model gives the correct digits. Accuracy is the share classified correctly.
Example input
Correct digit —
Prediction —
This image is held out. The network trains on other images.
Scroll sideways to see all ten outputs.