DELTAMEM: Incremental Experience Memory for LLM Agents via Residual Trees

Dimensions overall score 9.0

• {"file name": "input.pdf", "number of pages": 19, "author": "Haoran Tan; Zeyu Zhang; Zhicheng Cao; Rui Li; Xu Chen", "title": "DELTAMEM: Incremental Experience Memory for LLM Agents via Residual Trees", "creation date": null, "modification date": null, "kids": []}claim 82%
  Evidencepartial

  {"file name": "input.pdf", "number of pages": 19, "author": "Haoran Tan; Zeyu Zhang; Zhicheng Cao; Rui Li; Xu Chen", "title": "DELTAMEM: Incremental Experience Memory for LLM Agents via Residual Trees"

  Implicationmissing

  Implication not extracted yet.

  Verificationpartial

  partial

• DeltaMem organizes experience memory into two independent residual trees: one for goal-conditioned task experience as reusable skills and another for scene-level environment knowledge.method 95%
  Evidencepartial

  We propose DeltaMem, a framework that organizes experience memory into two independent residual trees, one storing goal-conditioned task experience as reusable skills and another for scene-level environment knowledge.

  Implicationpartial

  Directly stated in the abstract with clear description of the two-tree structure.

  Verificationpartial

  partial

• DeltaMem uses a failure-penalized similarity scan for retrieval, which locates the best match and reconstructs the full experience via root-to-match chain composition.method 90%
  Evidencepartial

  For retrieval, a failure-penalized similarity scan locates the best match, reconstructing the full experience via root-to-match chain composition.

  Implicationpartial

  Directly stated in the abstract with specific retrieval mechanism details.

  Verificationpartial

  partial

• DeltaMem consistently outperforms existing baselines across diverse interactive environments.result 85%
  Evidencepartial

  Experiments across diverse interactive environments show that DeltaMem consistently outperforms existing baselines.

  Implicationpartial

  Stated in the abstract as a result, but specific performance numbers are not provided in the excerpt.

  Verificationpartial

  partial

• DeltaMem includes an autonomous consolidation mechanism that distills high-frequency paths into new root nodes, enabling self-organization from general heuristics to specialized variants.method 90%
  Evidencepartial

  An autonomous consolidation mechanism distills high-frequency paths into new root nodes, enabling the trees to self-organize from general heuristics to specialized variants.

  Implicationpartial

  Directly stated in the abstract with clear description of the mechanism.

  Verificationpartial

  partial

• Storing experiences as independent, flat units leads to substantial redundancy and retrieval conflicts in LLM-based agents.limitation 80%
  Evidencepartial

  However, storing experiences as independent, flat units leads to substantial redundancy and retrieval conflicts, as similar episodes repeat overlapping content and subtle scene variations cause retrieved memories to offer contradictory guidance.

  Implicationpartial

  Stated as a problem in the abstract, but not directly measured in the excerpt.

  Verificationpartial

  partial

• DeltaMem uses residual experience, where newly acquired experience is an incremental variation of existing knowledge.technical 90%
  Evidencepartial

  To address this, we introduce residual experience, positing that newly acquired experience is often an incremental variation of existing knowledge.

  Implicationpartial

  Directly stated in the abstract as a core concept.

  Verificationpartial

  partial

• The code for DeltaMem is publicly available at https://github.com/import-myself/DeltaMem.market 100%
  Evidencepartial

  To facilitate future research, we release the code at https://github.com/import-myself/DeltaMem.

  Implicationpartial

  Explicitly stated in the abstract with the URL.

  Verificationpartial

  partial

• DeltaMem organizes experience memory into two independent residual trees: one for goal-conditioned task experience as reusable skills and another for scene-level environment knowledge.method 95%
  Evidencepartial

  We propose DeltaMem, a framework that organizes experience memory into two independent residual trees, one storing goal-conditioned task experience as reusable skills and another for scene-level environment knowledge.

  Implicationpartial

  Directly stated in the abstract with clear description of the two trees.

  Verificationpartial

  partial

• DeltaMem uses a failure-penalized similarity scan for retrieval, which locates the best match and reconstructs the full experience via root-to-match chain composition.method 95%
  Evidencepartial

  For retrieval, a failure-penalized similarity scan locates the best match, reconstructing the full experience via root-to-match chain composition.

  Implicationpartial

  Directly stated in the abstract.

  Verificationpartial

  partial

• DeltaMem consistently outperforms existing baselines across diverse interactive environments.result 90%
  Evidencepartial

  Experiments across diverse interactive environments show that DeltaMem consistently outperforms existing baselines.

  Implicationpartial

  Directly stated in the abstract, but specific performance numbers are not provided in the excerpt.

  Verificationpartial

  partial

• Storing experiences as independent, flat units leads to substantial redundancy and retrieval conflicts.technical 90%
  Evidencepartial

  However, storing experiences as independent, flat units leads to substantial redundancy and retrieval conflicts, as similar episodes repeat overlapping content and subtle scene variations cause retrieved memories to offer contradictory guidance.

  Implicationpartial

  Directly stated as a problem that DeltaMem addresses.

  Verificationpartial

  partial

Competitive landscape

DeltaMem is a novel memory framework for LLM agents that organizes experiences into residual trees for efficient incremental learning and retrieval.