Similar to the category, but not the exemplars: A study of generalization

Abstract

Reference point approaches have dominated the study of categorization for decades by explaining classification learning in terms of similarity to stored exemplars or averages of exemplars. The most successful reference point models are firmly grounded in the associative learning tradition-treating categorization as a stimulus generalization process based on inverse exponential distance in psychological space augmented by a dimensional selective attention mechanism. We present experiments that pose a significant challenge to popular reference point accounts which explain categorization in terms of stimulus generalization from exemplars, prototypes, or adaptive clusters. DIVA, a similarity-based alternative to the reference point framework, provides a successful account of the human data. These findings suggest that a successful psychology of categorization may need to look beyond stimulus generalization and toward a view of category learning as the induction of a richer model of the data.

A Cognitive Category-Learning Model of Rule Abstraction, Attention Learning, and Contextual Modulation

“…Thus, the diverse distributions of human learning success in Type II may stem from participants relying on different cognitive processes to master the task, but that rule instructions motivate the learner to engage in processes that trigger contextual modulation. This idea is corroborated by a recent study, in which the Type II task, which can also be described as an 'Exclusive Or' (XOR) problem, was extended to explicitly test rule abstraction or 'extrapolation' behavior (Conaway & Kurtz, 2017). In this study, participants were trained on a two-dimensional version of the problem (Figure 1B).…”

“…We then present simulations of CAL's behavior when learning the six problem types, and show how changes in its generalization and contrasting parameter (γ) explain the revised ordinal pattern of learning success (reduced Type II learning), which can be observed if the task instructions less strongly provoke rule abstraction (Kurtz et al, 2013) 15 . We then show, how this explanation relates to an account of (spontaneous) rule extrapolation in partial Exclusive-Or (XOR; or incomplete Type II) problems, as observed by Conaway and Kurtz (2017).…”

“…Despite the evident structural similarity between the classic Type II problem and its incomplete variant, ALCOVE (Kruschke, 1992), by its definition as an exemplar-similarity model, cannot predict this pattern of extrapolation. While there are other models that predict extrapolation in this task to some extent, such as the DIVergent Autoencoder model (DIVA; Kurtz, 2007; see further Conaway & Kurtz, 2017), it seems to be an open question whether there is an account that can simultaneously explain the Conaway and Kurtz (2017) result, and the various Type I-VI results (see also Kurtz et al, 2013; for further discussions). We argue that a mixture of rule learning (contextual modulation) and memorization might explain all these behavioral patterns, as well as several other puzzling empirical phenomena that we will discuss later.…”

“…Specifically, although the participants in Conaway and Kurtz (2017) never saw members of the category 'B' in the lower-right quadrant of the stimulus space during training (see Figure 15A), some participants still extrapolated 'B' in a later test phase (31% and 45% of participants in Exp. 1 and 2, respectively).…”

Creating Something Different: Similarity, Contrast, and Representativeness in Categorization

“…We used an artificial stimulus design: A two dimensional domain of squares, varying in color and size (see Figure 3a). These dimensions have been used in numerous classification learning studies (e.g., Conaway & Kurtz, 2016a, 2016bNosofsky, Gluck, Palmeri, & McKinley, 1994;Shepard et al, 1961). Unlike those used in the Jern and Kemp (2013) experiments, distance on these physical dimensions aligns more directly with perceptual similarity, allowing us to evaluate the role of contrast in categorization more precisely.…”