What is a multilayer perceptron (MLP)?

A multilayer perceptron (MLP) is a feedforward neural network made of an input layer, one or more hidden layers, and an output layer.

Each neuron in one layer is usually connected to every neuron in the next layer, which is why MLPs are also commonly described as fully connected neural networks.

Basic structure

A simple MLP looks like:

Input layer → Hidden layer(s) → Output layer

Each layer performs a weighted transformation followed by an activation function.

A neuron typically computes:

z = Wx + b

and then applies an activation:

a = f(z)

Why are hidden layers important?

Hidden layers allow the network to learn increasingly complex relationships between input features.

With nonlinear activation functions, an MLP can model patterns that a single linear transformation cannot represent.

How does an MLP learn?

During training:

• Inputs move forward through the network.
• The model produces a prediction.
• A loss function measures the prediction error.
• Backpropagation calculates gradients.
• An optimizer updates the weights and biases.

This process is repeated over many training examples.

What are MLPs used for?

MLPs can be used for tasks such as:

• Classification
• Regression
• Learning from tabular data
• Combining learned features before a final prediction

They are less specialized than architectures such as CNNs for images or RNNs for sequential data, but they remain one of the fundamental neural network architectures.

For the broader concept, see “What is a feedforward neural network?” and “What is a fully connected layer?”.

In short, an MLP is a fully connected feedforward neural network with one or more hidden layers that learns nonlinear relationships between inputs and outputs.