Multilayer Context Reasoning in a Neurobiologically Inspired Working Memory Model for Cognitive Robots
- Arthur Williams, Center for Computational Science, Middle Tennessee State University, Murfreesboro, Tennessee, United States
- Joshua Phillips, Dept. of Computer Science, Middle Tennessee State University, Murfreesboro, Tennessee, United States
AbstractThe brain's working memory system relies heavily on the mesolimbic dopamine system and the delivery of reward signals. The Working Memory Toolkit (WMtk) is a framework that incorporates working memory into robotic/artificial systems. The HWMtk is built on top of WMtk by using holographic reduced representations for concept encoding. This abstraction made it easier for end users to adopt the framework. While the HWMtk captures human and animal performance on some cognitive tasks, tasks with multiple context layers are still problematic. We extended the HWMtk framework by adding a multilayer context reasoning working memory system. We tested our system on the AX-CPT task and 1-2-AX-CPT task. Our results show that our model is capable of learning after a reasonable number of episodes and trials, thus making it amenable for comparison with human and animal performance data.
Return to previous page