Stability and Adaptation of Neural Networks
Abstract
Focused on unsupervised learning and adaptive fuzzy systems. Explored the differential competitive learning (DCL) law. We successfully benchmarked DCL against supervised competitive learning for phoneme recognition and centroid estimation. Proved structural stability for general competitive learning laws. Developed product-space clustering to develop adaptive fuzzy systems, which grow structured fuzzy rules from training data without supervision. Successfully benchmarked adaptive fuzzy systems against neural-network truck-and-trailer systems and Kalman-filter control systems for realtime target tracking.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 02, 1990
- Accession Number
- ADA230108
Entities
People
- Bart Kosko
Organizations
- University of Southern California