Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

Hierarchical Graph Memory for LLM Agents with Path-level Localization and Rewrite

Source

arXiv

arxiv.org/abs/2608.05095v1

Read the source

Source

Plain-Language Brief

The authors propose HiGram, an evolving hierarchical graph memory framework that aims to improve the efficiency and effectiveness of long-term reasoning agents by reducing irrelevant information during retrieval and improving evidence selection. The method includes a hierarchical graph memory organization, path-level localization using MicroGraphs, and a coordinated rewriting method for intra-unit memory and inter-unit dependencies.

What the Document Says

  • The paper presents HiGram, an evolving hierarchical graph memory framework for long-term reasoning agents.
  • HiGram organizes memory into a coarse-to-fine architecture of upper-level nodes and MemoryUnits to reduce irrelevant information during retrieval.
  • Path-level localization is achieved using query and update conditioned MicroGraphs.
  • A coordinated rewriting method is proposed for intra-unit memory and inter-unit dependencies.
  • Experiments show improvements in answer quality, token efficiency, answer accuracy, and query-valid evidence selection under dynamic, static, and conditional conflicts.

Key Concepts Mentioned

  • Long-term reasoning agents
  • Graph memory
  • Hierarchical graph
  • MemoryUnits
  • MicroGraphs
  • Path-level localization
  • Coordinated rewriting
  • Answer quality
  • Token efficiency
  • Dynamic conflicts
  • Static conflicts
  • Conditional conflicts

Claims Made by the Document

  • HiGram improves the efficiency and effectiveness of long-term reasoning agents.
  • Hierarchical graph memory reduces irrelevant information during retrieval.
  • Path-level localization using MicroGraphs is achieved.
  • Coordinated rewriting method improves intra-unit memory and inter-unit dependencies.
  • Experiments show improvements in answer quality, token efficiency, answer accuracy, and query-valid evidence selection.

Evidence Used

  • The authors state that their proposed method demonstrates substantial improvements over baselines.

Methods / Approach

  • Evolving hierarchical graph memory with path-level localization and rewriting
  • Hierarchical graph memory organization
  • MicroGraph-based path-level localization
  • Coordinated rewriting method

Limitations Stated by the Authors

  • Not stated in source

Limitations Noted by the Agent

Agent interpretation — not from the source.

  • The study is based on experiments and the actual performance may vary in real-world applications.
  • The paper does not provide a detailed comparison with other methods, which limits the understanding of its advantages and disadvantages.

Open Questions

  • What are the specific improvements in answer quality, token efficiency, and query-valid evidence selection compared to baselines?
  • How does HiGram perform under different types of conflicts and how does it compare to other methods?

Notes for Future Q&A

May include interpretation. Labeled where it does.

  • The paper focuses on long-term conversational question answering and conflict-aware memory evaluation.
  • The improvements in answer quality, token efficiency, and query-valid evidence selection are demonstrated through experiments but specific numbers are not provided.

Raw Abstract / Excerpt

Agents for long term reasoning require a memory that can be efficiently and effectively updated over time, as new facts and external feedback continue to arrive. Recently, graph memory has been adopted to offer structural organization for multi-hop retrieval and reasoning. However, existing methods store all memories in a flat graph, and accumulated historical memories can introduce irrelevant contexts and increase the cost of evidence selection during retrieval. Moreover, they typically update memory units independently, requiring repeated unit-wise rewrite to cover related changes. To address these issues, we propose HiGram, an evolving hierarchical graph memory framework with path-level localization and rewriting. Specifically, we first propose a hierarchical graph memory, which organizes the memory into coarse-to-fine architecture composed of upper-level nodes and MemoryUnits, thereby reducing the amount of irrelevant information during retrieval. We further propose MicroGraph-based path-level localization, which leverages query and update conditioned MicroGraphs to identify support subgraph and evidence path before rewrite. Finally, we propose a coordinated rewriting method that jointly revises intra-unit memory and inter-unit dependencies, enable valid dependency structures updating in the localized evidence path. Experiments on benchmarks for long-term conversational question answering and conflict-aware memory evaluation demonstrate that our method demonstrate substantial improvements over baselines in answer quality and token efficiency. Besides, our method improves answer accuracy and query-valid evidence selection under dynamic, static, and conditional conflicts.