Browse all publications by topic
Browse all publications by year
- M. Hasenjäger,
H. Ritter, and K. Obermayer. Active Data Selection for Fuzzy Topographic
Mapping of Proximities.
.
In Brewka G., Der R., Gottwald S., and Schierwagen A., editors,
Fuzzy-Neuro Systems 1999 - Computational Intelligence, pages
93-104, 1999.
(FTP Gzipped PostScript, 11 pages, 78 kb)
We investigate algorithms for the grouping and for the topographic
mapping of data items based on their mutual proximities. Given a full
proximity matrix we first show that a large portion of its entries can be
discarded without reducing the quality of the clustering solution. Although
the clustering procedure is quite sensitive to the omission of large and
small dissimilarity values, it is robust w.r.t. a deletion of up to 50 of
the medium sized entries. These facts are then further exploited by actively
selecting dissimilarity values for learning that are supposed to be most
relevant. We describe a seleciton strategy which is based on maximizing the
expected value of sample information. The advantage gained from active data
selection depends on problem size as well as on the strength with which
topography is imposed.
|
