Learning XOR

x1, x2→z1, z2→h1, h2→z3→y
Two hidden-neuron XOR network Inputs x1 and x2 feed two hidden preactivations z1 and z2, then hidden activations h1 and h2 feed output preactivation z3 and sigmoid prediction y. x1 input x2 input z1 h1 hidden 1 z2 h2 hidden 2 z3 y / sigmoid output
Active example (0, 0) → 0 Drag vertically ยท move slowly or hold Shift for fine control Threshold 0.500
XOR rows and every layer output
x1 x2 y pred t result BCE

BCE = -(t log y + (1 - t) log(1 - y))

Mean BCE
0.000
Accuracy
0%