BARTMAP vs. Iterative Two-Way Clustering, and Other Innovations

Abstract

The unsupervised nature of many clustering algorithms is one of their key advantages, even though many design decisions must still be made. In discussing some of these design decisions, we will briefly survey advantages and design issues in hierarchical clustering. Similarly, we will review some properties of Adaptive Resonance in engineering applications of clustering. This will lead us to an innovative application of an ART-inspired architecture (BARTMAP) to biclustering, an unsupervised version of heteroassociative learning. Comparison of this approach to other biclustering and traditional clustering approaches illustrates the advantages of biclustering in general and BARTMAP in particular. This advantage is further extended by the development of a new, hierarchical version of BARTMAP.

Meeting Name

Graduate Research Seminar, University of Missouri--Columbia (Mizzou IEEE CIS) (2012: May 8, Columbia, MO)

Department(s)

Electrical and Computer Engineering

Document Type

Article - Conference proceedings

Document Version

Citation

File Type

text

Language(s)

English

Publication Date

01 May 2012

Link to Full Text

Share

 
COinS