Concepts

There are two types of AI's. Predictive and Generetive AI

Predictive

Regression

The model output is a single continuous numeric value, such as the price of a house, sea level, or temperature.

Classification

The model output is a category or a label. There are multiple types of classification problems:

Binary: a yes/no problem. The model outputs a single Boolean value. Spam detection is a classic example of binary classification.

Multiclass: the model outputs a single value that can take more than two discrete values. For example, in digit recognition, the output value corresponds to 10 classes (0-9).

Multilabel: the model outputs multiple discrete values. Identifying the music genres of a song is an example of this type.

Clustering

The model outcome is a set of clusters, or groups of cases. Each case is located within a cluster. The output also includes information such as the center of each cluster and the number of elements. Clustering problems are typically solved with unsupervised learning.

Generative AI (GenAI)

GenAI is focused on generating high-quality, realistic data, including text, images, audio, video, or code. Modern GenAI applications and models, such as Granite, LLama, ChatGPT, or Stable Diffusion, have recently emerged as industry-transforming technologies.

In GenAI, the target variables are a vector, or a matrix of continuous values. These values represent the generated content, such as the pixels of a new image. Generation problems often require powerful and large models, such as autoencoder or transformer neural networks, and the use of unsupervised learning is common.

Reinforcement AI

In reinforcement AI, the model produces actions that an agent or a system must execute within a given environment, taking into account the state of the environment and the rewards or penalties received as a result of previous actions. This type of AI is mostly based on reinforcement learning.