Overview

Feedforward networks turn raw predictors into learned non-linear features, making them the core approximation engine behind many quant forecasting and classification models. This topic starts with the MLP architecture, then makes its prediction interface precise through output layers and losses, before deriving the backward computations that fit every parameter jointly.

By the end of this note you should be able to:

• describe an MLP and explain why depth requires non-linear activations;
• select an output layer and loss for regression or classification;
• execute and derive the main pieces of vectorised backpropagation.

1. Contents
2. The big picture