Affective Replay for Reinforcement Learning

This article introduces Affective Replay with M2MR Integration, a novel experience replay mechanism for reinforcement learning that combines emotion-driven prioritization with synthetic memory generation. Unlike traditional experience replay methods that rely on static priorities or uniform sampling, our hybrid approach dynamically assesses experience relevance through emotion-based weighting while augmenting the replay buffer with high-confidence synthetic experiences generated from emotionally similar memories. The approach is motivated by the challenges of applying reinforcement learning in financial environments characterized by regime shifts and non-stationary data. Evaluated on market-based simulations using real S&P 500 and VIX data, Affective Replay achieves consistent and statistically significant performance improvements over all baselines, including combined experience replay, memory-to-memory replay (M2MR), variance-aware continuous learning, and prioritized experience replay. Our results suggest that improving how reinforcement learning systems prioritize and reuse historical observations may enhance model adaptability in financial applications.