A real convolutional network trains on synthetic 8×8 digit-like patterns. The Conv→ReLU→MaxPool→Dense pipeline runs entirely in the browser.

Predictions

Each tile shows an input pattern with its true label (top-left) and predicted label (bottom-right). Green border = correct.

Training accuracy

Network: Conv(3×3, 4 filters) → ReLU → MaxPool(2×2) → Dense(16) → Dense(4).