How and Why Does Category Learning Cause Categorical Perception?

Abstract

Learning to categorize requires distinguishing category members from non-members by detecting the features that covary with membership. Human subjects were trained to sort visual textures into two categories by trial and error with corrective feedback. Difficulty levels were increased by decreasing the proportion of covariant features. Pairwise similarity judgments were tested before and after category learning. Three effects were observed: (1) The lower the proportion of covariant features, the more trials it took to learn the category and the fewer the subjects who succeeded in learning it. After training, (2) perceived pairwise distance increased between categories and, to a lesser extent, (3) decreased within categories, at all levels of difficulty, but only for successful learners. This perceived between-category separation and within-category compression is called categorical perception (CP). A very simple neural network model for category learning using uniform binary (0/1) features showed similar CP effects. CP may occur because learning to selectively detect covariant features and ignore non-covariant features reduces the dimensionality of perceived similarity space. In addition to (1) – (3), the nets showed (4) a strong negative correlation between the proportion of covariant features and the size of the CP effect. This correlation was not evident in the human subjects, probably because, unlike the formal binary features of the input to the nets, which were all uniform, the visual features of the human inputs varied in difficulty.

Integrating unsupervised and reinforcement learning in human categorical perception: A computational model

“…The work [ 23 ] proposes a computational model of categorical perception with which the authors investigate different learning processes. They adopt a functional approach based on machine learning, including an auto-encoder (AE; [ 26 ]) and a classifier.…”

“…These models take different approaches, as they focus on the interaction between low-level and high-level information at different neuronal sites (e.g. apical and basal dendrites; [ 19 ]), systems supporting speech production [ 20 ], self-organising mechanisms [ 21 ], visual competitive hierarchies [ 22 ], and effects of supervised signals [ 23 ]. Other models of CP investigate Bayesian inferential mechanisms [ 24 ] and embodied evolutionary influences [ 25 ].…”

Disentangling the effects of attentional weighting and associative mediation in perceptual learning reveals no evidence for associative mediation.

“…de Leeuw et al (2016) reported a similar effect of task difficultly on acquired equivalence and distinctiveness effects using more complex stimuli. Conversely, Pérez-Gay et al (2017) did not find an effect of stimulus discriminability on categorical perception effects, but their stimuli were textures generated by tessellation of simple checkerboard patterns, and it is arguable that even their most discriminable stimuli were substantially more similar than any of the stimuli used by Hall et al (2003) or by Meeter et al (2009). One explanation for this effect of discriminability might be found in the suggestion made by Sutherland and Mackintosh (1971) that attention to very salient stimuli might change more slowly than to less salient stimuli, a principle that has been incorporated into other models of associative learning (e.g., Suret & McLaren, 2003, 2005).…”

Effects of Inductive Learning and Gamification on News Veracity Discernment

“…One such factor is that Bad News only presents examples of one category (i.e., fake news) rather than two categories (i.e., true and fake news). Most research that has successfully improved discernment between two categories has done so by presenting examples of both categories (Higgins & Ross, 2011;Pérez-Gay et al, 2017). This allows for within-and between-category comparisons, which have both been found to be indispensable for category learning (Hammer et al, 2008).…”