As within, so without, as above, so below: Common mechanisms can support between- and within-trial category learning dynamics.

Abstract

Two fundamental difficulties when learning novel categories are deciding (a) what information is relevant and (b) when to use that information. Although previous theories have specified how observers learn to attend to relevant dimensions over time, those theories have largely remained silent about how attention should be allocated on a within-trial basis, which dimensions of information should be sampled, and how the temporal order of information sampling influences learning. Here, we use the adaptive attention representation model (AARM) to demonstrate that a common set of mechanisms can be used to specify: (a) How the distribution of attention is updated between trials over the course of learning and (b) how attention dynamically shifts among dimensions within a trial. We validate our proposed set of mechanisms by comparing AARM’s predictions to observed behavior in four case studies, which collectively encompass different theoretical aspects of selective attention. We use both eye-tracking and choice response data to provide a stringent test of how attention and decision processes dynamically interact during category learning. Specifically, how does attention to selected stimulus dimensions gives rise to decision dynamics, and in turn, how do decision dynamics influence which dimensions are attended to via gaze fixations?

Cognitive Inertia: Cyclical Interactions Between Attention and Memory Shape Learning

“…One computational model we discuss here is the adaptive attention representation model (AARM; Galdo et al, 2022; Turner, 2019; Weichart, Evans, et al, 2022; Weichart, Galdo, et al, 2022), which is derived from other exemplar-based models of categorization (Estes, 1986; Medin & Schaffer, 1978; Nosofsky, 1986). In AARM (as well as other models in this class), attention is represented as a vector containing the amount of attention for each dimension of information.…”

“…As attention increases for one particular dimension, it enhances that dimension’s contribution to the category response as well as the probability of fixating to that dimension during the decision period. When we make explicit the connection between fixation probability and strength of encoding ( Weichart, Galdo, et al, 2022 ), AARM expresses differential encoding such that features that were thought to be relevant at one time point in the learning sequence are more strongly encoded in memory and will have a larger impact on both decisions and encoding in the future. One of AARM’s central characteristics is its commitment to understanding how attention is deployed on a trial-by-trial basis as learners interact with their environment.…”

“…Each location in the space corresponds to a different possible allocation of attention, and yellow-green areas correspond to regions of the space that satisfy the goal(s) used in each panel. We now highlight three potential goals that can independently, or in combination, direct attention during learning (Galdo et al, 2022).…”

“…Although these two components of decision-making—selective attention and memory—traditionally separate into different aspects of cognition, there has been some theoretical interest in embracing the idea that they are inexorably intertwined. The extant theories detailing attention-memory interactions can roughly be distinguished on the basis of how attention is allocated to previous episodic events (i.e., exemplars; Aha & Goldstone, 1992; Nelson & Shiffrin, 2013), clusters of related episodic events (Braunlich & Love, 2022; Love et al, 2004), or individual features of those previous events (Carvalho & Goldstone, 2022; Turner et al, n.d.; Weichart, Galdo, et al, 2022). All of these theories provide evidence that selective attention affects the way the stimulus environment is represented in memory, and the evidence stems from applying computational models to response data from carefully constructed categorization experiments as well as data from neuroimaging (Mack et al, 2013, 2016, 2020) and from eye tracking (Blair et al, 2009; Galdo et al, 2022; Kruschke & Blair, 2000; Rehder & Hoffman, 2005a, 2005b; Weichart, Galdo, et al, 2022).…”

Dynamic Retrieval of Events and Associations from Memory: An Integrated Account of Item and Associative Recognition

“…Moreover, the dynamics of orienting attention to memory representations are analogous to the dynamics of orienting attention to a target element in an external stimulus (Logan et al, 2022). Attention can also shift rapidly within a trial in order to perform categorization tasks (Weichart et al, 2022). Finally, rapid within-trial shifts of attention have been shown to be able to account for the extra-list feature effect (Mewhort & Johns, 2000) in recognition memory (Osth et al, In press).…”