Core ACT-R & Cognitive Architectures

The present paper describes an implementation of contextual memory and a basic event-handler for the ACT-Up cognitive architecture which maintains its scalability and appropriateness for rapid-prototyping while adding essential features and lowering the barrier to entry for new users. This includes describing a theory-neutral implementation of working memory and spreading activation, in addition to a basic associative learning mechanism. An example of rapid prototyping for algorithm development is presented using the serial memory task described in Klein, Addis, and Kahana (2025). This study describes how contiguity effects change across sequential list presentations across three serial and free recall conditions. We further describe how to use generative AI and the event handler to automatically create cognitive experiments directly from the Methods section of research papers.

This is an in-person presentation on July 19, 2026 (09:00 ~ 09:20 EDT).

Category learning constitutes a fundamental cognitive process in humans. While a considerable body of research has studied this through passive reception of stimuli, real-world scenarios often require learners to actively select the information they learn from. An approach to category learning involving active and passive hypothesis testing is employed to compare learning by reception (using a preselected sequence of stimuli) and selection (using self-chosen stimuli). For this, a cognitive plausible ACT-R model is presented with perception and motor abilities and without memorization of past stimuli. The ACT-R model is evaluated against a subset of a dataset from human participants and a Bayesian model on a rule-based and information-integration category learning task. In the rulebased learning task, the ACT-R model adequately captured the learning curves of human participants. However, in the information-integration task, the accuracy of the model was an average of 9.6 percentage points lower than human performance, a discrepancy primarily driven by the simplified hypothesis structure in the model. The ACT-R model demonstrates how a simple cognitive strategy can solve a category learning task without relying on the memorization of stimuli by employing hidden states, such as a confidence measure of hypotheses.

This is an in-person presentation on July 19, 2026 (09:20 ~ 09:40 EDT).

To interpret and understand the visual environment, humans do not encode each stimulus independently but organize stimuli into structured groups. We extended the ACT-R architecture with a mechanism that represents perceptual clusters derived from visual features. A controller implemented in Python analyzed RGBA information obtained from a task display in the Unity game engine and generated clusters from visual stimuli. These cluster representations were then transmitted to a Common Lisp component and converted into a form accessible to the ACT-R visual module. The extended architecture was evaluated using a simple recognition task. A clustering model developed with this architecture produced behavior consistent with the pattern expected when humans rely on cluster-based representations. Comparisons with human data showed that the basic model without the clustering mechanism more closely matched human performance in simpler conditions where judgments could be made based on individual stimulus retrieval. In contrast, the clustering model’s performance was closer to human performance in conditions that required greater cognitive resources, such as integrating information from multiple stimuli. The proposed extension provides a foundation for ACT-R models to interact with complex visual scenes derived from realistic task environments.

This is an in-person presentation on July 19, 2026 (09:40 ~ 10:00 EDT).