A cognitive category-learning model of rule abstraction, attention learning, and contextual modulation.

Abstract

We introduce the CAL model (Category Abstraction Learning), a cognitive framework formally describing category learning built on similarity-based generalization, dissimilarity-based abstraction, two attention learning mechanisms, error-driven knowledge structuring, and stimulus memorization. Our hypotheses draw on an array of empirical and theoretical insights connecting reinforcement and category learning. The key novelty of the model is its explanation of how rules are learned from scratch based on three central assumptions. (1) Category rules emerge from two processes of stimulus generalization (similarity) and its direct inverse (category contrast) on independent dimensions. (2) Two attention mechanisms guide learning by focusing on rules, or on the contexts in which they produce errors. (3) Knowing about these contexts inhibits executing the rule, without correcting it, and consequently leads to applying partial rules in different situations. The model is designed to capture both systematic and individual differences in a broad range of learning paradigms. We illustrate the model's explanatory scope by simulating several benchmarks, including the classic Six Problems, the 5-4 problem, and linear separability. Beyond the common approach of predicting average response probabilities, we also propose explanations for more recently studied phenomena that challenge existing learning accounts, regarding task instructions, individual differences in rule extrapolation in three different tasks, individual attention shifts to stimulus features during learning, and other phenomena. We discuss CAL's relation to different models, and its potential to measure the cognitive processes regarding attention, abstraction, error detection, and memorization from multiple psychological perspectives.

g-distance: On the comparison of model and human heterogeneity

“…the one where every point in parameter space is equiprobable, seems unlikely prima facie ; for example, most psychometric measures are not uniformly distributed. In cases where modelers specify individual variation in terms of hyper-parameters, these are typically the parameters of non-uniform (often Gaussian) distributions (see Daw, Gershman, Seymour, Dayan, & Dolan, 2011;and Schlegelmilch, Wills, & Helversen, 2022). Dealing with non-uniform distributions poses no insurmountable problems, beyond the fact that model designers seldom specify hyper-parameters for their models.…”

Multiple pathways to widespread fears: Disentangling idiosyncratic fear generalization mechanisms using computational modeling

“…Traditionally, researchers in the field have focused on the group gradient, which is the average of the individual response gradients 6 . Along with the summary statistics, numerous statistical models and theories were applied to quantitatively describe and predict generalization behavior [9][10][11][12][13][14][15] .…”

Individual differences in stimulus identification, rule induction, and generalization of learning.

“…While aggregate response patterns across a wide range of experiments in both animals and humans follow the shape predicted by a similarity-based generalization process (Ghirlanda & Enquist, 2003; Mednick & Freedman, 1960), this is far less the case for individuals, who exhibit a variety of response patterns (Lee et al, 2021; Zaman, Chalkia, et al, 2021). Identifying the causes responsible for this interindividual variability remains a challenge for many associative generalization models (but see Schlegelmilch et al, 2021). As overgeneralization of fear has been implicated in many forms of psychopathology (Dunsmoor & Paz, 2015; Kindt, 2014; Lissek & Grillon, 2010; Vlaeyen & Linton, 2012), the identification of the various underlying mechanisms yielding interindividual differences have important clinical, as well as theoretical implications.…”

“…Inspired by recent research in humans where response gradients in specific individuals deviated from a similarity-based response pattern, various authors have proposed the involvement of an inductive or cognitive process (Lee et al, 2018; Lovibond et al, 2020; Schlegelmilch et al, 2021; Wong & Lovibond, 2017). For instance, while some participants generalize along a green-blue stimulus dimension based on similarity of the test stimulus to the CS+, others reported adhering to a more abstract rule (i.e., the bluer the stimulus, the more I expect a shock).…”

“…These findings are reminiscent of the distinction between strategies based on exemplar similarity versus abstract rules made in category learning (e.g., Erickson & Kruschke, 1998; Little & McDaniel, 2015; Nosofsky, 1986), nonlinear compound discriminations (Shanks & Darby, 1998), multiple cue judgment (e.g., Hoffmann et al, 2016), and function learning (e.g., DeLosh et al, 1997), suggesting that these individual differences in response rules are general across a variety of inductive tasks. Currently, the sources of these individual differences in inductive rules in the context of associative learning are unknown (but see Schlegelmilch et al, 2021).…”