An MLP stacks multiple layers of neurons with nonlinear activations between them. This is the foundational deep learning architecture:

Python / PyTorch

import torch.nn as nn

class MLP(nn.Module):
    def __init__(self, input_dim, hidden_dim, output_dim):
        super().__init__()
        self.layers = nn.Sequential(
            nn.Linear(input_dim, hidden_dim),   # Hidden layer 1
            nn.ReLU(),                           # Nonlinearity
            nn.Linear(hidden_dim, hidden_dim),   # Hidden layer 2
            nn.ReLU(),
            nn.Linear(hidden_dim, output_dim),   # Output layer
        )
    def forward(self, x):
        return self.layers(x)

model = MLP(input_dim=784, hidden_dim=256, output_dim=10)

The Universal Approximation Theorem (Hornik, 1989) proves that an MLP with a single hidden layer and nonlinear activation can approximate any continuous function — given enough neurons. But "enough neurons" can be impractically many. Adding depth is often more efficient than adding width.