← Back to Topics

Autoencoder Backprop

Forward Pass

net_h = w1×x1 + w2×x2
h = sigmoid(net_h)
net_o = w3×h
o = sigmoid(net_o)

Backward Pass

δ_o = (target - o) × o×(1-o)
δ_h = δ_o × w3 × h×(1-h)

Weight Update

Δw = η × δ × input_to_weight
w_new = w_old + Δw + α×Δw_previous
View in LibraryTry in Sandbox