Understanding AI Hierarchy: ML, Deep Learning, Generative AI

Machine learning encompasses a wide range of techniques for building systems that improve from data. Within this field, deep learning stands out as a type of machine learning that uses artificial neural networks. These networks are inspired by the human brain and consist of many interconnected nodes, or neurons, that learn by processing data and making predictions. Deep learning models typically have many layers of neurons, which allows them to learn more complex patterns than traditional machine learning models. As one expert puts it, “deep learning is a type of machine learning that uses artificial neural networks allowing them to process more complex patterns than machine learning.”

Learning with Data

Neural networks can be trained with both labeled and unlabeled data. When a small amount of labeled data is combined with a large amount of unlabeled data, the approach is called semi‑supervised learning. In this setting, the labeled examples help the network learn basic concepts, while the unlabeled examples enable it to generalize to new situations. This dual use of data lets the model benefit from the strengths of supervised and unsupervised methods.

Generative AI’s Place

Generative AI is a subset of deep learning that also relies on artificial neural networks. It can process labeled and unlabeled data using supervised, unsupervised, and semi‑supervised methods. Large language models (LLMs) belong to the same deep‑learning family, making them another subset of deep learning. As a concise statement notes, “Generative AI is a subset of deep learning” and “Large language models are also a subset of deep learning.”

Types of Machine Learning Models

Deep‑learning and broader machine‑learning models can be divided into two categories: generative and discriminative.

• Discriminative models focus on classification or prediction. They are typically trained on labeled data and learn the relationship between data features and labels. In probabilistic terms, they model the conditional probability P(Y|X), which represents the probability of an output given an input. Once trained, a discriminative model predicts labels for new data points. As described, “a discriminative model is a type of model that is used to classify or predict labels for data points.”

• Generative models aim to create new data instances. They learn the joint probability distribution P(X, Y), capturing how inputs and outputs co‑occur. By understanding this distribution, a generative model can generate new content that resembles the training data. This is captured in the definition: “a generative model generates new data instances based on a learned probability distribution of existing data.”

Both model types rely on artificial neural networks, but they differ in what they learn and how they are applied.

Model Comparison Example

Consider an image‑recognition task involving dogs. A discriminative model learns P(Y|X) and can classify a given picture as “dog” rather than “cat.” In contrast, a generative model learns P(X, Y). After estimating the probability that an image is a dog, it can generate a new picture of a dog from scratch. This illustrates the visual distinction between traditional machine‑learning models, which discriminate between existing data instances, and generative AI models, which generate new data instances. As one summary states, “Generative models can generate new data instances and discriminative models discriminate between different kinds of data instances.”