A scalable reinforcement learning framework inspired by hippocampal memory mechanisms for efficient contextual and sequential decision making
Abstract Efficient decision-making in context-dependent, sequential tasks remains a fundamental challenge in reinforcement learning (RL). Inspired by the function of the brain’s hippocampal system, we introduce Hippocampal-Augmented Memory Integration (HAMI), a biologically inspired memory-based RL...
Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-07-01
|
| Series: | Scientific Reports |
| Online Access: | https://doi.org/10.1038/s41598-025-10586-x |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Efficient decision-making in context-dependent, sequential tasks remains a fundamental challenge in reinforcement learning (RL). Inspired by the function of the brain’s hippocampal system, we introduce Hippocampal-Augmented Memory Integration (HAMI), a biologically inspired memory-based RL framework that leverages symbolic indexing, hierarchical memory refinement, and structured episodic retrieval to enhance both learning efficiency and adaptability. We also propose Hierarchical Contextual Sequences (HiCoS), a structured RL environment grounded in neuroscience studies on episodic and sequence memory and context-driven decision-making, which serves as a controlled testbed for evaluating biologically inspired memory-based decision-making systems. Our experimental results demonstrate that HAMI achieves high decision accuracy and improved sample efficiency while maintaining low memory utilization. HAMI’s architecture exhibits significantly lower inference latency than baseline memory-based methods, and its structured retrieval is well-suited for further hardware acceleration with non-volatile memory (NVM)-based content-addressable memory (CAM). By integrating biologically inspired memory mechanisms with structured symbolic representations, HAMI provides a scalable and efficient memory-based RL framework for tackling context-dependent sequential tasks. |
|---|---|
| ISSN: | 2045-2322 |