The Quest for Simplicity in Human Learning: Identifying the Constraints on Attention

Abstract

For better or worse, humans live a resource constrained existence; only a fraction of the sensations our body experiences ever reach conscious awareness, and we store a shockingly small subset of these experiences in short-term memory for later use. Despite these observations, most theories of learning assume that, given feedback about a new experience, knowledge is updated so as to minimize subsequent errors with minimal consideration of cognitive capacity constraints. Acknowledging that human cognition has clear biological limitations, we explored the degree to which human learning could be better described with sets of biases toward simpler and more parsimonious mental representations (i.e., simplicity biases) relative to an error-driven, accuracy-maximizing normative model. Taking the normative model as a basis, we developed a suite of nested computational models that use various mechanistic simplicity biases to explain learning. We fit these models to four data sets that varied in the type of learning needed to achieve high accuracy. Across all data sets, we found consistent evidence that the best descriptors of human learning were models with mechanisms that instantiated a constrained optimization process, where errors were minimized subject to constraints on both attention and memory. Importantly, whereas normative models failed to account for patterns of attentional deployment over time, models with simplicity biases accounted well for both choice responses and fixation data as participants learned various categorization tasks.

Distributed neural systems enable flexible attention updating during category learning

“…Given that it is often the case that multiple dimensions provide similarly diagnostic information, the learner could conceivably seek to reduce time or effort spent on each individual trial by only attending to a subset of informative dimensions before making a response, with minimal detriment to overall accuracy. This idea has been supported by previous work, in which AARM with additional mechanisms to optimize for secondary computational goals outperformed a baseline unconstrained variant when fit to behavioral and eye-tracking data (Galdo et al, 2021). While a strict error-reduction policy for attention updating that is standard among contempo-rary adaptive attention models was sufficient for predicting accuracy across trials, accounting for individualized computational goals in the gradient specification was necessary for predicting trial-level information sampling behavior via eyetracking.…”

“…In previous work, support for AARM's mechanisms of attention allocation was provided by fits to simultaneous streams of choice and eye-tracking data that were collected while participants learned novel categories (Galdo et al, 2021). Across paradigms of varying complexity, AARM accurately predicted increases in accuracy that coincided with increased probability of selectively attending to goal-relevant dimensions, as measured by trial-level gaze fixations.…”

“…The observed relative learning rates across category types provide considerable empirical constraint that contemporary theories of category learning are expected to account for in order to be regarded as viable (e.g. Kruschke, 1992;Nosofsky et al, 1994;Goodman et al, 2008;Galdo et al, 2021).…”

“…As we get to know a new person, for example, we may be able to categorize their mood as happy, sad, or angry based on specific elements of their facial expression, tone of voice, or body language. In an effort to explain how humans can learn new categories quickly even when they are multivariate, probabilistic, or non-linearly separable, computational models of categorization formalize the processing stream that links memories of previous experiences to representations of new items (Nosofsky, 1986;Kruschke, 1992;Love et al, 2004;Galdo et al, 2021). Across contemporary models, dynamic allocation of selective attention to goal-relevant dimensions is often implicated as the critical mechanism through which categorization accuracy improves across trials.…”

“…The current study focuses specifically on the Adaptive Attention Representation Model (AARM; Galdo et al, 2021), an example of the class of adaptive attention models described above. The conceptual basis of AARM comes from context theory, which assumes previously-experienced items (i.e.…”