Cognitive Agents Integrating Rules and Reinforcement Learning for Context-Aware Decision Support

While context-awareness has been found to be effective for decision support in complex domains, most of such decision support systems are hard-coded, incurring significant development efforts. To ease the knowledge acquisition bottleneck, this paper presents a class of cognitive agents based on self-organizing neural model known as TD-FALCON that integrates rules and learning for supporting context-aware decision making. Besides the ability to incorporate a priori knowledge in the form of symbolic propositional rules, TD-FALCON performs reinforcement learning (RL), enabling knowledge refinement and expansion through the interaction with its environment. The efficacy of the developed Context-Aware Decision Support (CaDS) system is demonstrated through a case study of command and control in a virtual environment.