Integrating Reinforcement and Episodic Learning in a General Cognitive Architecture

点击翻译按钮获取中文摘要

This project aims to extend the Soar architecture to include episodic and reinforcement learning. Although Soar is a mature and widely known cognitive architecture, the extensions will enhance Soar by providing it an episodic memory of prior events that can improve future decision making and the ability to improve its behavior in response to rewards. This research will explore possible synergies between these learning mechanisms including using reinforcement learning for learning memory encoding and recall strategies, using internal episode-based models to boost reinforcement learning and using a model of emotions for affecting learning rates. Given that episodic memory is an important aspect of human cognition, this project will also strengthen Soar as a cognitive model. Besides building on on-going work on Soar, this project will build on results from closely allied fields such as Case-Based Reasoning.