A tiny feed-forward net learning XOR live โ watch neurons fire, weights update, and loss fall.
Controls
Edges = weights (thickness โ magnitude, blue + /
red โ). Moving dots are signals flowing forward;
the glow behind an edge shows the size of its gradient during back-prop.
Network โ forward pass & gradient flow
positive weight
negative weight
gradient glow
โ neuron activation
Loss (mean squared error)
Current lossโ
Epoch0
Total updates0
XOR truth table & predictions
xโ
xโ
target
output
error
Prediction cell is green when the rounded output
matches the target. The net has converged once all four rows are green and loss is near zero.
โโWโ โ