Several machine learning techniques (kth-nearest-neighbour, clustering) depend on distances between observations. When observations can be recorded as points in Euclidean space, Euclidean distance is the obvious thing to use. But how can we find distances between objects with discrete properties (like sex, or drug, or kind of adverse reaction)? And when we have properties on widely differing scales, possibly correlated, how can we combine distance measures on properties to give us distance measures on whole objects?
This talk presents two methods for defining distances given discrete properties, one of which even allows us to use k-means clustering. It also presents two methods for combining distances.
This is on-going research, so the talk will NOT be presenting results.
Last modified: Thursday, 28-Jul-2005 17:23:30 NZST
This page is maintained by the seminar list administrator.