Catalogue of Artificial Intelligence Techniques
Keywords: learning, unsupervised learning
Author(s): Jeremy Wyatt
In unsupervised learning the learner is given a set of input patterns, but does not receive a corresponding set of output patterns (as happens in Supervised Learning). It therefore does not have any explicit representation of desired input-output pairs. The aim of unsupervised learning is to find regularities in the input set. This often simply done by grouping together similar input patterns. Different learning procedures use different measures of similarity. They also produce different representations of the regularities found in the input set. These include Discrimination Nets and Feature Maps. Unsupervised learning procedures are often based on statistical techniques. Conceptual Clustering, for example, is related to cluster analysis and self-organising feature maps are related to principal components analysis. One principal difference between unsupervised learning and statistical analysis is that the latter relies on all the data being present at the same time, i.e., at the start of learning. Because many unsupervised learning techniques learn incrementally they can converge to different partitions of the input space given different sequences of input patterns. Significant work on unsupervised learning has been done by Kohonen, Grossberg, and Rumelhart and Zipser, among others.
No references to display.