Artificial Intelligence Research Group

Some Past and Current Projects
Under Development

13. Unconventional Distance Metrics
Investigator(s): Dr Weizhong Liu

This study focuses on developing unconventional methods for measuring the differences between items of data. Conventionally the Nearest Neighbour measures employed in Case Based Reasoning and other classification methods employ simple euclidian measures of distance to gauge the relative positions of data with respect to classification. Several approaches are being looked at: Background knowledge is being used to weight attributes so that the distance between datum is a function of both the Euclidian distance and the weighting. Supervised learning is being used to weight cases. Statistical methods are used to judge the utility of cases depending on their past performance and accuracy when used for retrieval. This knowledge is then incorporates in the distance metric. The thrust of this work is to find exemplar cases so that retrieved cases are closer to the exemplar cases that they would be if nearest neighbour techniques were used.

Liu, W.Z. & White, A.P. (1995). "A comparison of nearest neighbour and tree-based discriminant analysis." Journal of Statistical and Computational Simulation, 53, pp. 41-50.

Next