OnARTMAP: A Fuzzy ARTMAP-based Architecture.

摘要: Abstract Fuzzy ARTMAP (FAM) copes with the stability–plasticity dilemma by adaptive resonance theory (ART). Despite such an advantage, suffers from a category proliferation problem, which leads to high number of categories and decrease in performance for unseen patterns. Such drawbacks are mainly caused overlapping region (noise) between classes. To overcome these drawbacks, we propose ARTMAP-based architecture robust noise, named OnARTMAP, both online batch learning. Our neural networks (OnARTMAP 1 OnARTMAP 2 ) proposed learning have two-stage process, while our network o incremental has just single iterative process. Two new modules proposed, detection module (ORDM) another one similar A R T , called c . The ORDM finds categories, computes stores special areas. In proposal, weights ordinary estimated data outside area. An alternative second stage strategy is presented focuses on improving generalization performance. On basis achievements, can infer that improve categories. proposals were applied artificial real datasets, as well compared several counterparts (Fuzzy ARTMAP, ART-EMAP, μ BARTMAP).