Research map / Large language models / Retrieval and knowledge
AssoMem: Scalable Memory QA with Multi-Signal Associative Retrieval
Kai Zhang, Xinyuan Zhang, Ejaz Ahmed, Hongda Jiang, Caleb Kumar, Kai Sun, Zhaojiang Lin, Sanat Sharma, Shereen Oraby, AARON COLAK, Ahmed A Aly, Anuj Kumar and 2 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add temporal signal to the ranker (ablation) | AssoMem · LongMemEval_m | Recall@6: 73.39 -> 80.87 | Improved |
| add importance (PPR) signal to the ranker (ablation) | AssoMem · LongMemEval_m | Recall@6: 75.81 -> 80.87 | Improved |
| add clue nodes to the memory graph (ablation) | AssoMem · LongMemEval_m | Recall@6: 79.75 -> 80.87 | Improved |
| add clue nodes to the memory graph (ablation) | AssoMem · LongMemEval_m | Acc@10: 72.51 -> 69.17 | Hurt |
| use MI-driven weight assignment instead of fixed weights (ablation) | AssoMem · LongMemEval_m | Recall@6: 76.79 -> 80.87 | Improved |
| use mutual information fusion instead of learnable fusion (logistic regression) (ablation) | AssoMem · LongMemEval_m | Recall@6: 72.92 -> 82.64 | Improved |
| use mutual information fusion instead of information gain (ablation) | AssoMem · LongMemEval_m | Recall@6: 80.94 -> 82.64 | Improved |
| fine-tune generator with denoising multi-task data (ablation) | Qwen2.5-32B-Instruct · LongMemEval_m | Accuracy: 64.72 -> 73.88 | Improved |
| fine-tune generator with denoising multi-task data (ablation) | LlaMA3.2-3B-Instruct · LongMemEval_m | Accuracy: 26.91 -> 33.43 | Improved |
| use AssoMem retrieval instead of topic grouping | best baseline · LongMemEval_m | Recall@10: 79.14 -> 84.96 | Improved |
| use AssoMem retrieval instead of topic grouping at larger memory | best baseline · LongMemEval_l | Recall@10: 62.29 -> 69.33 | Improved |
| use hybrid multi-granularity retrieval instead of utterance-flat | retrieval · LongMemEval_m | Recall@10: 70.18 -> 78.97 | Improved |
Closest papers by meaning
- From Isolated Conversations to Hierarchical Schemas: Dynamic Tree Memory Representation for LLMs (ICLR 2025 · similarity 0.87)
- LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (ICLR 2025 · similarity 0.87)
- HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models (NeurIPS 2024 · similarity 0.86)
- PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents (ICML 2026 · similarity 0.86)
- From RAG to Memory: Non-Parametric Continual Learning for Large Language Models (ICML 2025 · similarity 0.86)
- SeCom: On Memory Construction and Retrieval for Personalized Conversational Agents (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
