Feed-forward network

In deep learning, a feed-forward network (or multilayer perceptron, MLP) is a type of neural network that aims to approximate a function $f^{\ast }$. For example, if we have a function $f^{\ast }:\vec{x}\to y$, an MLP defines a mapping $f:\vec{x},\vec{\theta }\to \vec{y}$ where $\theta$ are parameters that the model learns to best approximate the function.

Feed-forward models get their name because information flows from the input $\vec{x}$ to the output $\vec{y}$ without any feedback connection. Recurrent neural networks are feed-forward models that are extended to include feedback.

Note that MLPs are limited often because they require we pre-process the input. This makes it unideal for vision tasks, where convolutional neural networks don’t require this step.