Artificial neural networks: S. Grossberg's adaptive resonance theory

Journal/Book: Cesk Psychol. 1997; 41: Publ House Czech Acad Sci, Vodickova 40, Prague 1, Czech Republic 112 29. Academia. 360-369.

Abstract: Grossberg's Adaptive Resonance Theory (ART) represents a mathematical model of nervous processes. In it's technical realization on a computer it is a classification device working entirely - independently on trainer's interferences. The theory was developed on grounds of psychological phenomenon that our perception is matched against our expectations. The basis of ART is formed by an attentional subsystem consisting of two processing fields: information is processed from a bottom perception field to a top cognition field where a top-down processes are initiated. Bottom-up processes are learning processes developing a category of an input pattern (category in the top field) and top-down processes represent a recall of the prototype (learned expectation represented by the category). If an input pattern and the prototype are similar enough then these two fields dynamicaly influence each other - the bottom-up signal initiates the top-down one, which again initiates the bottom-up, etc. The notion of adaptive resonance refers to such a state. In ART only resonant percepts are conscious. The orienting subsystem supports recognition of new inputs and reorganizes memory. By means of ART theory Grossberg explains the rise of intention and abstraction, hypothesis testing, visual and auditory perception and ilusions, etc.

Note: Article

Keyword(s): artificial neural networks; modelling of psychological processes; adaptive resonance theory; PATTERN-CLASSIFICATION; FEEDBACK; BRAIN

Zurück | Weiter