ML//neural network//loss function

- A single number: how wrong the model is.

A single number: how wrong the model is.

Backprop minimizes this — every weight update exists to make this number smaller.

Cross-entropy for classification, MSE for regression.