How and why is working memory (WM) capacity limited? Traditional cognitive accounts focus either on limitations on the number or items that can be stored (slots models), or loss of precision with increasing load (resource models). Here, we show that a neural network model of prefrontal cortex and basal ganglia can learn to reuse the same prefrontal populations to store multiple items, leading to resource-like constraints within a slot-like system, and inducing a trade-off between quantity and precision of information. Such ‘chunking’ strategies are adapted as a function of reinforcement learning and WM task demands, mimicking human performance and normative models. Moreover, adaptive performance requires a dynamic range of dopaminergic signals to adjust striatal gating policies, providing a new interpretation of WM difficulties in patient populations such as Parkinson’s disease, ADHD, and schizophrenia. These simulations also suggest a computational rather than anatomical limit to WM capacity.

Evidence accumulation models (EAMs) are the dominant framework for modeling response time (RT) data from speeded decision-making tasks. While providing a good quantitative description of RT data in terms of abstract perceptual representations, EAMs do not explain how the visual system extracts these representations in the first place. To address this limitation, we introduce the visual accumulator model (VAM), in which convolutional neural network models of visual processing and traditional EAMs are jointly fitted to trial-level RTs and raw (pixel-space) visual stimuli from individual subjects in a unified Bayesian framework. Models fitted to large-scale cognitive training data from a stylized flanker task captured individual differences in congruency effects, RTs, and accuracy. We find evidence that the selection of task-relevant information occurs through the orthogonalization of relevant and irrelevant representations, demonstrating how our framework can be used to relate visual representations to behavioral outputs. Together, our work provides a probabilistic framework for both constraining neural network models of vision with behavioral data and studying how the visual system extracts representations that guide decisions.