ICCM: Cognitive Architectures

This paper presents Xyrast, an integrative model of human response processes on the Raven's Matrices family of fluid intelligence tests, and reports on a simulation study addressing its response characteristics and verisimilitude. Xyrast is implemented in the Clarion cognitive architecture and models the influence of response strategy, working memory capacity, and persistence on performance. Simulations suggest that the model captures a wide range of phenomena offering, in some cases, novel explanations for observed results. These findings suggest several avenues for future research.

This is an in-person presentation on July 20, 2023 (15:20 ~ 15:40 UTC).

When we learn new tasks, rather than starting from scratch, we often reuse skills that we have learned previously. By integrating these previously learned skills in a new way, we can learn how to do new tasks with little effort. In this research, we test a method aimed at identifying the skills reused between tasks. More specifically, we use a knowledge graph as a tool for identifying reused skills. From this knowledge graph, we built a cognitive model that shows how the identified skills can be integrated to solve a new task. The final cognitive model can successfully solve a variety of related but distinct tasks. This shows it is possible to use knowledge graphs to identify the skills reused between tasks. This ability may benefit how we approach learning. Knowing, in advance, the skills needed to successfully complete a new task may allow us to learn said task in an easier, more focused manner.

This is an in-person presentation on July 20, 2023 (15:40 ~ 16:00 UTC).

Likelihood functions form the basis for statistical inference techniques, including maximum likelihood estimation, and Bayesian estimation/model comparison. Unfortunately, deriving likelihood functions analytically for cognitive architectures such as ACT-R can be challenging, if not impossible in some cases, often requiring considerable time and expertise. Simulation-based approximations are computationally intensive, making them impractical to implement in real-time applications. We demonstrate how recently developed techniques for learning intractable likelihood functions with neural networks can be applied to a visual search model based on ACT-R, and reused once trained. Our work extends prior applications in two ways: (1) we demonstrate that the technique can be scaled to a large number of conditions based on the size of the visual search array, and (2) we demonstrate that the technique is applicable to both unimodal and multimodal versions of the model. We conclude with a discussion for scaling up neural network techniques for approximating likelihood functions.

This is an in-person presentation on July 20, 2023 (16:00 ~ 16:20 UTC).