Hippocampal Replay Mechanisms for Adaptive Memory Optimization: A Sleep-Inspired AI Framework

Modern artificial intelligence (AI) systems are predominantly built upon artificial neural networks (ANNs) that, despite recent advancements, remain limited in key areas such as energy efficiency, adaptability, and alignment with biological processes. Traditional ANNs rely on computationally intensive and rigid learning mechanisms, often lacking the flexibility required for dynamic, real-world environments. One crucial biological process that has received limited attention in AI research is sleep particularly the memory consolidation that occurs through interactions between the hippocampus and cortex. Growing evidence highlights the significance of offline mechanisms such as bidirectional replay, pattern separation, and pattern completion during sleep. These processes support robust memory consolidation and contribute to adaptive behaviour in biological systems. In this work, we propose a novel AI framework inspired by hippocampal memory consolidation, with a specific focus on path optimization tasks through forward replay mechanisms. Experiments conducted on a simulated environment demonstrate the effectiveness of the proposed approach, while maintaining low energy consumption.