![[2602.13933] HyMem: Hybrid Memory Architecture with Dynamic Retrieval Scheduling](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.13933] HyMem: Hybrid Memory Architecture with Dynamic Retrieval Scheduling
Summary
The paper presents HyMem, a hybrid memory architecture designed to enhance the performance of large language models (LLMs) in extended dialogues by optimizing memory management through dynamic retrieval scheduling.
Why It Matters
HyMem addresses the critical challenge of memory efficiency in LLMs, which often struggle with complex reasoning in lengthy interactions. By improving memory management, this architecture could significantly enhance AI applications in natural language processing and other fields, making them more effective and resource-efficient.
Key Takeaways
- HyMem introduces a dual-granular storage scheme for improved memory efficiency.
- Dynamic retrieval scheduling allows for better adaptation to complex queries.
- The architecture reduces computational costs by 92.6% while maintaining performance.
- HyMem outperforms existing models on LOCOMO and LongMemEval benchmarks.
- The approach is inspired by cognitive principles, aiming for a more human-like memory management system.
Computer Science > Artificial Intelligence arXiv:2602.13933 (cs) [Submitted on 15 Feb 2026] Title:HyMem: Hybrid Memory Architecture with Dynamic Retrieval Scheduling Authors:Xiaochen Zhao, Kaikai Wang, Xiaowen Zhang, Chen Yao, Aili Wang View a PDF of the paper titled HyMem: Hybrid Memory Architecture with Dynamic Retrieval Scheduling, by Xiaochen Zhao and 4 other authors View PDF HTML (experimental) Abstract:Large language model (LLM) agents demonstrate strong performance in short-text contexts but often underperform in extended dialogues due to inefficient memory management. Existing approaches face a fundamental trade-off between efficiency and effectiveness: memory compression risks losing critical details required for complex reasoning, while retaining raw text introduces unnecessary computational overhead for simple queries. The crux lies in the limitations of monolithic memory representations and static retrieval mechanisms, which fail to emulate the flexible and proactive memory scheduling capabilities observed in humans, thus struggling to adapt to diverse problem scenarios. Inspired by the principle of cognitive economy, we propose HyMem, a hybrid memory architecture that enables dynamic on-demand scheduling through multi-granular memory representations. HyMem adopts a dual-granular storage scheme paired with a dynamic two-tier retrieval system: a lightweight module constructs summary-level context for efficient response generation, while an LLM-based deep module ...