Among Supervised Learning and Reinforcement Learning (RL), Unsupervised Learning is one of the primary neural network learning paradigms. In this paradigm, the computer is given unlabeled data, leaving it on its own to find structure in its input. Unsupervised learning can be used to discover hidden patterns in data without any explicit guidance or instruction.
These models use self-learning algorithms, so they infer their own rules and then structure the information based on patterns. These algorithms are suitable for complex processing tasks like organizing large datasets into clusters.
The three main uses for unsupervised learning are:
- Clustering: An algorithms that explores raw unlabeled data and breaks into down into clusters based on similarities or differences. See Clustering, K-Means Clustering, DBSCAN Clustering, etc.
- Association: A rule-based approach to reveal relationships between data points in large datasets. Essentially, it discovers correlations and co-occurrences within the data. See Apriori, Eclat, FP-Growth Algorithms, etc
- Dimensionality Reduction: A technique to reduce the numbers of features (dimensions) in a dataset. While more data is generally better for ML, it can also be hard to visualize. Dimensionality extracts important features from the dataset and reduces the number of irrelevant features present.