Loss Functions & Gradients

Implement mean squared error, mean absolute error, and the MSE gradient.

Implement binary and categorical cross-entropy with numerical stability — the standard classification loss.

Implement finite-difference gradient checking — the standard way to verify backprop implementations.