Unsupervised learning ( English unsupervised learning ) called machine learning without known in advance target values, without reward by the environment. The (learning) machine tries to recognize patterns in the input data that deviate from structureless noise. An artificial neural network is based on the similarity to the input values and adapts the weights accordingly. Different things can be learned. Automatic segmentation ( clustering ) or compression of data to reduce dimensions are popular.
A very simplified example: Imagine different fruits (apples, pears, strawberries, oranges) that are all in a common basket. The basket therefore contains the amount of data to be "segmented". Now one fruit is to be taken out at random. Then look for similarities with the fruits already on the ground. When something suitable has been found, the fruit should be added. If not, put them somewhere where there is space. This has to be continued until all fruits have been "segmented" according to their properties (appearance, smell, color, taste, etc.). There are now various piles of fruits on the floor, sometimes larger, smaller or the same depending on the frequency of occurrence. In practical terms, these are the clusters .
The attempt is made to represent many input values in a more compact form, whereby as little information as possible should be lost. The principal component analysis may be, for example, understood as the compression method, when the least significant components of the data is omitted.
This is practically the same as a linear auto-encoder ; this is a multi-layered artificial neural network , the target values of which are the input values, whereby a hidden layer with fewer nodes as input values serves as a “bottleneck”. The activations of these neurons are the compressed data from which the original data are then to be reconstructed (decompressed) as well as possible.
- Geoffrey Hinton , Terrence J. Sejnowski (Eds.): Unsupervised Learning: Foundations of Neural Computation . MIT Press, 1999, ISBN 0-262-58168-X (English).
- Richard O. Duda , Peter E. Hart , David G. Stork : Pattern classification (2nd edition) . Wiley, New York 2001, ISBN 0-471-05669-3 , Unsupervised Learning and Clustering, pp. 571 (English).