Human Category Learning: Implications for Backpropagation Models

• Published in: Connection science Vol. 5; no. 1; pp. 3 - 36
• Main Author: KRUSCHKE, JOHN K.
• Format: Journal Article
• Language: English
• Published: London Taylor & Francis Group 01.01.1993; Taylor & Francis
• Subjects: Backpropagation; Biological and medical sciences; catastrophic forgetting; categorization; coarse coding; condensation; dimensional attention; error-driven learning; filtration; Fundamental and applied biological sciences. Psychology; Learning. Memory; linear boundaries; local representation; Psychology. Psychoanalysis. Psychiatry; Psychology. Psychophysiology; Human; Forgetting; Backpropagation; Computer simulation; Memory; Error; Cognition; Representation; Neural network; Experimental study; Algorithm; Connectionism; Linear model; Learning; Selective attention; Acquisition process; Categorization
• ISSN: 0954-0091; 1360-0494
• DOI: 10.1080/09540099308915683

Backpropagation (Rumelhart et al., 1986) was proposed as a general learning algorithm for multi-layer perceptrons. This article demonstrates that a standard version of backprop fails to attend selectively to input dimensions in the same way as humans, suffers catastrophic forgetting of previously learned associations when novel exemplars are trained, and can be overly sensitive to linear category boundaries. Another connectionist model, ALCOVE (Kruschke 1990, 1992), does not suffer those failures. Previous researchers identified these problems; the present article reports quantitative fits of the models to new human learning data. ALCOVE can be functionally approximated by a network that uses linear-sigmoid hidden nodes, like standard backprop. It is argued that models of human category learning should incorporate quasi-local representations and dimensional attention learning, as well as error-driven learning, to address simultaneously all three phenomena.