- Rating
- 1
- Confidence
- 3
- Informativeness
- 1
- Presentation
- 2
- Soundness
- 1
Read review
Paper summary
This paper studies how a model should access a fixed long-term conversation memory when answering questions. The authors propose REMIND, where the model rewrites the query, retrieves memories, and decides to search again or answer based on the evidence it got. The policy is trained with GRPO using rewards for answer, retrieval, query quality, new evidence and repeated queries.
Critical strengths
-
The problem is meaningful. The paper points out that the most useful memory is not always the most similar one, which is a fair observation.
-
The method is described clearly. The full reward formulas and the algorithm are given, and the code is released, which helps reproduction.
-
REMIND improves over the same-backbone Base RAG at both 4B and 8B scale.
Critical weaknesses
-
The novelty is limited, and the method does not really solve its own motivation. The paper claims relevance comes from causal and temporal relations, but it still uses fixed similarity retrieval with query rewriting; nothing handles time validity or causality.
-
The experiments has serious problems. Common memory baselines like Mem0, A-Mem and MemoryOS are not compared at all. I think the paper is clearly below the bar of ICLR.
-
Important ablations are missing. There is no prompt-only iterative retrieval baseline, and no ablation on each reward term, so it is unclear where the improvement actually comes from.
-
Training data comes from LoCoMo and LongMemEval, and evaluation also uses these two datasets, but how they are split is never explained. MemoryAgentBench is not reported separately either.
Questions for authors
-
Could the authors analyze how the method actually handles causal and temporal relevance?
-
Could the authors add comparisons with common memory baselines such as Mem0, A-Mem and MemoryOS?
-
Could the authors add a prompt-only iterative retrieval baseline without RL training, and an ablation that removes each reward term one by one?
-
Could the authors explain how the training and evaluation data from LoCoMo and LongMemEval are split and whether they overlap, and report MemoryAgentBench results separately?