Unifying Multiple Knowledge Domains Using the ARTMAP Information Fusion System
Carpenter, Gail A.
MetadataShow full item record
Sensors working at different times, locations, and scales, and experts with different goals, languages, and situations, may produce apparently inconsistent image labels that are reconciled by their implicit underlying relationships. Even when such relationships are unknown to the user, an ARTMAP information fusion system discovers a hierarchical knowledge structure for a labeled dataset. The present paper addresses the problem of integrating two or more independent knowledge hierarchies based on the same low-level classes. The new system fuses independent domains into a unified knowledge structure, discovering cross-domain rules in this process. The system infers multi-level relationships among groups of output classes, without any supervised labeling of these relationships. In order to self-organize its expert system, ARTMAP information fusion system features distributed code representations that exploit the neural network’s capacity for one-to-many learning. The fusion system software and testbed datasets are available from http://cns.bu.edu/techlab
RightsCopyright 2008 Boston University. Permission to copy without fee all or part of this material is granted provided that: 1. The copies are not made or distributed for direct commercial advantage; 2. the report title, author, document number, and release date appear, and notice is given that copying is by permission of BOSTON UNIVERSITY TRUSTEES. To copy otherwise, or to republish, requires a fee and / or special permission.