Proceedings of the Annual Meeting of the Cognitive Science Society

The brain’s working memory system relies heavily on themesolimbic dopamine system and the delivery of reward sig-nals. The interaction between the prefrontal cortex (PFC) andthe basal ganglia are the main components simulated in work-ing memory models. The Working Memory Toolkit (WMtk) isa framework that allows the incorporation of working memoryinto robotic/artificial systems. The HWMtk is built on top ofWMtk by using holographic reduced representations for con-cept encoding. This allows end users to adopt the frameworkwithout the need to understand details of the algorithms in-volved. While the HWMtk captures human and animal per-formance on some cognitive tasks, tasks with multiple con-text layers are still problematic. We extended the HWMtkframework by adding a multilayer context reasoning work-ing memory system. We tested our system on the AX-CPTtask, 1-2-AX-CPT task and a 2-layer context task that is par-tially observable. Our results show that our model is capableof learning after a reasonable number of trials, thus making itamenable for comparison with human and animal performancedata.