The theoretical underpinnings of mass-based dissimilarities
Abstract
A generic definition of mass-based dissimilarity has established that this data dependent dissimilarity is better than the commonly used (data independent) distance measure in three data mining tasks: clustering, classification and anomaly detection. This project pushesthe research frontier to unearth the theoretical underpinnings of mass-based dissimilarity through 1) comparative analyses of existing data dependent (metric) dissimilarities and the mass-based (nonmetric) dissimilarity, 2) formal analyses of the conditions in which distancemeasure will perform poorly in the three above tasks, and 3) a new mass-based maximum margin classifier which enhances mass-based dissimilarity through learning.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 28, 2017
- Source ID
- FA23861714034
Entities
People
- Kai Ming Ting
Organizations
- Air Force Office of Scientific Research
- Federation University Australia
- United States Air Force