What Is a Deep Neural Network

News coverage often describes large language models as AI models. That is true, but the vocabulary can quickly become tangled. A large language model is also a type of deep neural network, and PyTorch, the library this course is about, is a deep learning library. So a single system can be described with several overlapping labels at once. This article untangles those labels by treating the three words as nested circles rather than as three unrelated topics.

The three terms form a containment chain. Artificial intelligence is the outermost circle. Inside it sits machine learning, and inside machine learning sits deep learning. Each inner circle inherits everything that is true of the circles around it. A deep learning system is therefore also a machine learning system and also an AI system.

The definition is deliberately broad. It describes a goal and a family of capabilities rather than a single technique. Any computer system that handles language, spots patterns, or chooses between options in a way that ordinarily takes human judgment falls under this umbrella. The source also notes that AI is still far from achieving general intelligence at the human level. In practice, the field describes systems that do specific tasks well, not systems that match people in every respect.

The phrase without being explicitly programmed is the key to machine learning. In ordinary programming, a person writes the rules that the computer follows. In machine learning, the algorithm works out useful patterns from data, and it can improve as it receives more data and feedback. Machine learning is a subfield of AI for exactly this reason.

This shift is why machine learning moved AI beyond strict rule-based systems. A rule-based system can only handle what its author anticipated. A learning system can adapt to new inputs or changing environments instead of relying only on rules written in advance.

Machine learning is not a side branch of AI. The source describes it as integral to the evolution of AI, and it is the technology behind many advances visible today, including large language models. Recommendation systems used by online retailers and streaming services, email spam filtering, voice recognition in virtual assistants, and self-driving cars all appear in the source as machine learning applications.

Deep learning is a subcategory of machine learning that trains and applies deep neural networks. The word deep does not mean difficult or advanced in a vague sense. The source is specific: the deep in deep learning refers to the multiple hidden layers of artificial neurons or nodes.

Hidden here means the layers that sit between the data going in and the result coming out. Data enters the network at one end, passes through a series of intermediate layers, and produces a result at the other end. The depth of the network is a property of how many of these intermediate layers it has. These layers are what let a network model complex, nonlinear relationships in data.

Deep learning is particularly well suited to unstructured data such as images, audio, and text, and this is the main reason it underpins large language models. The source contrasts this with traditional machine learning techniques, which excel at simple pattern recognition. Unstructured data does not arrive as neat rows and fixed columns, so the layered structure of a deep network is a good match for it.