Unsupervised learning is a type of machine learning in which the program uses trial-and-error methods to determine the best way of accomplishing the task at hand. By experimenting, AI programs can learn to classify symptoms and reach a diagnosis for the medical community, or troubleshoot a problem for technical support.
Unsupervised Learning and Clusters
In order to improve diagnostics, unsupervised learning is used to teach an AI program how to classify data into groups called clusters. During the training process, the program must learn how the data is best sorted. In order to achieve this, the program is provided with data to classify, in one of several ways.
- In some cases the data is in one large group, and the program is given the correct number of subgroups. The program then searches for a method of organization that produces the desired result.
- In other cases, the data is in many small clusters, and the program is required to re-classify the clusters into larger clusters.
- AI programs can also learn by breaking down large clusters into smaller clusters.
In all cases, the association of items into various groups can take a number of forms. Each set of data that the program receives will fine-tune its ability to classify the data as required to achieve the desired results.
Artificial intelligence programs that train themselves through unsupervised learning can work with limited human oversight to provide an accurate diagnosis for disease or electronic failure. By learning to sort symptoms and errors into the appropriate clusters, the program can determine the underlying cause of the problems.
- Example One: A patient reports a list of symptoms. The AI program, trained to classify symptoms into clusters, recognizes that the majority of those symptoms belong to the cluster of data that is associated with one particular disease. This results in a diagnosis of the disease for the patient.
- Example Two: A computer’s error messages are run through a diagnostics program. The AI program used to troubleshoot the error learned using the same methodology, but different classification criteria. This program also sorts the data received in order to determine the best category into which to fit the errors. Once the errors have been placed in the proper categories, the cause becomes clear. This also allows the program to offer solutions based on the reason for the error.
Clustering as a Learning Tool
Clustering is a learning tool that allows artificial intelligence programs to accurately diagnose disease as well as to isolate problems arising from electronic errors. There are many ways for an AI program to learn to function, but the use of unsupervised machine learning reduces programming time drastically. This type of training can be time-consuming, however, as the program must run through each potential result. The program’s accuracy continues to improve through operation, without the requirement for any further programming, which balances the initial investment of time.
Bozdogan, H.; Statistical Data Mining and Knowledge Discovery. Chapman and Hall. (2003).
Cios, K., Kurgan, L., Pedrycz, W., Swiniarski, R. Data Mining: A Knowledge Discovery Approach. Published 2007, by Springer US.
Leondes, C. Expert systems: The technology of knowledge management and decision making for the 21st century, Volume 1. Harcourt. (2002)
Originally published on Suite101.
Decoding Science. One article at a time.