Deep learning is a branch of machine learning that trains Neural Networks with many layers — typically dozens to hundreds — stacked between the input and output. Each layer learns increasingly abstract representations: early layers detect low-level features like edges or character n-grams, while later layers combine these into high-level concepts like objects, syntax, or semantic meaning. The term "deep" refers to this depth of representation, not to any philosophical sophistication of the approach.

Think of it like an assembly line where raw parts pass through many workstations, each one adding a more refined level of processing. A sheet of metal enters at one end, and a finished engine component exits at the other — no single station could have produced it alone.

A minimal deep learning model in PyTorch illustrates the stacked-layer structure:

import torch.nn as nn

model = nn.Sequential(
    nn.Linear(784, 512),
    nn.ReLU(),
    nn.Linear(512, 256),
    nn.ReLU(),
    nn.Linear(256, 10),
)

Each Linear layer followed by an Activation Function like ReLU forms one processing stage; stacking them is what makes the network "deep."

Deep learning is enabled by Gradient Descent, which propagates error signals backward through all layers to update every Parameter simultaneously. The Transformer is currently the dominant deep learning architecture for language, while the Convolutional Neural Network dominated vision tasks before it. Foundation Models are large deep learning systems pretrained on internet-scale data.

You will encounter deep learning as the underlying engine behind nearly every modern AI capability, from image recognition and speech transcription to code generation and protein structure prediction.