![[2602.14038] Choosing How to Remember: Adaptive Memory Structures for LLM Agents](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.14038] Choosing How to Remember: Adaptive Memory Structures for LLM Agents
Summary
The paper presents FluxMem, a novel framework for adaptive memory structures in large language model (LLM) agents, addressing limitations of existing memory systems by enabling context-adaptive memory selection.
Why It Matters
As LLMs become integral in various applications, enhancing their memory capabilities is crucial for maintaining coherent interactions over time. FluxMem's adaptive approach could significantly improve performance in tasks requiring long-term memory, making it relevant for AI development and research.
Key Takeaways
- FluxMem allows LLM agents to utilize multiple memory structures for better performance.
- The framework adapts memory selection based on interaction-level features.
- Improvements of 9.18% and 6.14% were observed in benchmark tests.
- A three-level memory hierarchy enhances long-horizon memory evolution.
- The use of a Beta Mixture Model improves memory fusion processes.
Computer Science > Artificial Intelligence arXiv:2602.14038 (cs) [Submitted on 15 Feb 2026] Title:Choosing How to Remember: Adaptive Memory Structures for LLM Agents Authors:Mingfei Lu, Mengjia Wu, Feng Liu, Jiawei Xu, Weikai Li, Haoyang Wang, Zhengdong Hu, Ying Ding, Yizhou Sun, Jie Lu, Yi Zhang View a PDF of the paper titled Choosing How to Remember: Adaptive Memory Structures for LLM Agents, by Mingfei Lu and 10 other authors View PDF HTML (experimental) Abstract:Memory is critical for enabling large language model (LLM) based agents to maintain coherent behavior over long-horizon interactions. However, existing agent memory systems suffer from two key gaps: they rely on a one-size-fits-all memory structure and do not model memory structure selection as a context-adaptive decision, limiting their ability to handle heterogeneous interaction patterns and resulting in suboptimal performance. We propose a unified framework, FluxMem, that enables adaptive memory organization for LLM agents. Our framework equips agents with multiple complementary memory structures. It explicitly learns to select among these structures based on interaction-level features, using offline supervision derived from downstream response quality and memory utilization. To support robust long-horizon memory evolution, we further introduce a three-level memory hierarchy and a Beta Mixture Model-based probabilistic gate for distribution-aware memory fusion, replacing brittle similarity thresholds. Experi...
