Multi-Layer Perceptron

Adjust weights manually or train with gradient descent to match a target wave

Output vs Target

Blue = network prediction. Green dashed = target. Adjust weights or train to make them match.

Predicted

Target sin(x)

Training points

Network

Click a connection to edit its weight. Click a neuron to edit its bias.

Architecture:

— 0.00

Blue connections = positive weight · Red = negative · Thickness = magnitude

Training

Gradient descent minimizes loss by nudging each weight in the direction that reduces error.

Learning rate:

Iterations:

Epoch: 0/150 Loss: —

Each iteration: forward pass predicts output → MSE loss measures error → backpropagation computes each weight's gradient → weights update by −(learning rate × gradient).