continuous-time agents

Self-Monitoring Benefits from Structural Integration: Lessons from Metacognition in Continuous-Time Multi-Timescale Agents

This research investigates the utility of self-monitoring capabilities (metacognition, self-prediction) in reinforcement learning agents, finding they offer no significant benefit. The implemented modules collapsed to near-constant outputs, indicating the ineffectiveness of the tested mechanisms.