Abstract: How do we know when a reinforcement learning policy needs to adapt? In non13 stationary environments, agents must adapt and learn in environments that 14 change dynamically. We propose a finite-horizon model-free solution using a hier15 archical learning structure with fuzzy systems. The higher-level learning policy 16 advises the lower-level policy when to start and stop learning based on the tempo17 ral differences calculated within the lower-level. Major differences in the temporal 18 difference of each action produced by an agent may indicate environment change. 19 This structure is tested with multi-agent differential games in both the cooper20 ative and competitive aspect. Results show that this method is quick to notice 21 and adapt the policy within relatively few learning episodes.
PDF
Neurocomputing
Keywords: reinforcement learning, non-stationary environment, multi-agent system, fuzzy systems.