Chapter 7: Other types of machine learning tasks
Anomaly detection is a specific type of clustering where one is trying to detect whether a data point is noise or not. This is used to check whether a credit card transaction is fraudulent, or for instance, if a machine needs repair.
Though it could be supervised, the main concept behind anomaly detection is that anomalies are rare or have yet to occur.
Dimensionality reduction is a task that transforms data into a lower dimension. It could, for instance, transform images into a compressed format, or help visualize, otherwise high-dimensional, datasets.
Association-rule learning tries to find relationships between events. For instance, it tries to infer that customers who buy product A, also buy product B, or that people who visit website A and website B, will also visit website C.
Reinforcement learning is the task of learning appropriate behaviour within an environment. For instance, how to get out of a maze, or how to navigate a car through traffic. It is often referred to as “learning with a critic” because often, the optimal behaviour is not known, and therefore cannot be taught. Instead, we are able to tell how good or how bad a decision was. Learning then becomes trying to receive as much positive feedback, and as little negative feedback, as possible