{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:T5TAHX3BRVAQGEPVQ32ZLBHQAA","short_pith_number":"pith:T5TAHX3B","schema_version":"1.0","canonical_sha256":"9f6603df618d410311f586f59584f000182d65814519a5232ba48889ea3049bf","source":{"kind":"arxiv","id":"2502.14802","version":2},"attestation_state":"computed","paper":{"title":"From RAG to Memory: Non-Parametric Continual Learning for Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"HippoRAG 2 enhances Personalized PageRank with deeper passage integration and online LLM use to outperform standard RAG on factual, sense-making, and associative memory tasks.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bernal Jim\\'enez Guti\\'errez, Sizhe Zhou, Weijian Qi, Yiheng Shu, Yu Su","submitted_at":"2025-02-20T18:26:02Z","abstract_excerpt":"Our ability to continuously acquire, organize, and leverage knowledge is a key feature of human intelligence that AI systems must approximate to unlock their full potential. Given the challenges in continual learning with large language models (LLMs), retrieval-augmented generation (RAG) has become the dominant way to introduce new information. However, its reliance on vector retrieval hinders its ability to mimic the dynamic and interconnected nature of human long-term memory. Recent RAG approaches augment vector embeddings with various structures like knowledge graphs to address some of thes"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2502.14802","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-02-20T18:26:02Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"328069aa3c1ce500a395303508bcebfdd7e0f9d727dec094984ce4c47707d8eb","abstract_canon_sha256":"f654d7b685b2fe89159830b4af05295abf0236d8455be59c85157462e84296a5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:45.966818Z","signature_b64":"fhyqW5UPODucyI4g0pq3bCvYsNb2L29ZGLLIxiDVa2EJ7fjn+dRDIa+qEj0jOZmirQGgXL8OG9OzV/ZXpdEnBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9f6603df618d410311f586f59584f000182d65814519a5232ba48889ea3049bf","last_reissued_at":"2026-05-17T23:38:45.966151Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:45.966151Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From RAG to Memory: Non-Parametric Continual Learning for Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"HippoRAG 2 enhances Personalized PageRank with deeper passage integration and online LLM use to outperform standard RAG on factual, sense-making, and associative memory tasks.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bernal Jim\\'enez Guti\\'errez, Sizhe Zhou, Weijian Qi, Yiheng Shu, Yu Su","submitted_at":"2025-02-20T18:26:02Z","abstract_excerpt":"Our ability to continuously acquire, organize, and leverage knowledge is a key feature of human intelligence that AI systems must approximate to unlock their full potential. Given the challenges in continual learning with large language models (LLMs), retrieval-augmented generation (RAG) has become the dominant way to introduce new information. However, its reliance on vector retrieval hinders its ability to mimic the dynamic and interconnected nature of human long-term memory. Recent RAG approaches augment vector embeddings with various structures like knowledge graphs to address some of thes"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"HippoRAG 2 ... achieving a 7% improvement in associative memory tasks over the state-of-the-art embedding model while also exhibiting superior factual knowledge and sense-making memory capabilities.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the specific enhancements to passage integration depth and online LLM usage in the Personalized PageRank process are what drive the reported gains and that the chosen memory-task benchmarks accurately reflect the dynamic, interconnected nature of human long-term memory.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"HippoRAG 2 improves on standard RAG and prior HippoRAG by adding deeper passage integration and more effective LLM use in Personalized PageRank, delivering superior performance on factual, sense-making, and associative memory tasks including a 7% gain in associative memory over state-of-the-art.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"HippoRAG 2 enhances Personalized PageRank with deeper passage integration and online LLM use to outperform standard RAG on factual, sense-making, and associative memory tasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d52b94b5686174a7751a3dabc92a899852c9f5914f2dc35988f0da3f0170f3a4"},"source":{"id":"2502.14802","kind":"arxiv","version":2},"verdict":{"id":"41a68707-565c-42db-a3a7-6d945d3fdc3f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T01:30:40.665831Z","strongest_claim":"HippoRAG 2 ... achieving a 7% improvement in associative memory tasks over the state-of-the-art embedding model while also exhibiting superior factual knowledge and sense-making memory capabilities.","one_line_summary":"HippoRAG 2 improves on standard RAG and prior HippoRAG by adding deeper passage integration and more effective LLM use in Personalized PageRank, delivering superior performance on factual, sense-making, and associative memory tasks including a 7% gain in associative memory over state-of-the-art.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the specific enhancements to passage integration depth and online LLM usage in the Personalized PageRank process are what drive the reported gains and that the chosen memory-task benchmarks accurately reflect the dynamic, interconnected nature of human long-term memory.","pith_extraction_headline":"HippoRAG 2 enhances Personalized PageRank with deeper passage integration and online LLM use to outperform standard RAG on factual, sense-making, and associative memory tasks."},"references":{"count":11,"sample":[{"doi":"","year":2024,"title":"LightRAG: Simple and Fast Retrieval-Augmented Generation","work_id":"6118de04-b8eb-4163-826a-0f91a1bcdf14","ref_index":1,"cited_arxiv_id":"2410.05779","is_internal_anchor":true},{"doi":"10.1145/511446.511513","year":2002,"title":"In: Proceedings of the 11th international conference on World Wide Web (WWW)","work_id":"abdf368e-802b-4a1b-9c71-c31818bf4ff9","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.18653/v1/2024.acl-long.77","year":2023,"title":"Mitigating catastrophic forgetting in large language models with self-synthesized rehearsal","work_id":"0a00fc04-f30a-45f4-8121-b456df382bab","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.18653/v1/2024","year":2024,"title":"Towards General Text Embeddings with Multi-stage Contrastive Learning","work_id":"861a61de-66fe-49d1-b1ab-11f8b082a4cc","ref_index":4,"cited_arxiv_id":"2308.03281","is_internal_anchor":true},{"doi":"10.18653/v1/2022.emnlp-main","year":2022,"title":"Red Teaming Language Models with Language Models.Proceedings of EMNLP 2022, pp","work_id":"3601bd7c-4994-45a6-8587-7fc9c9aa1540","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":11,"snapshot_sha256":"31aefd0bc4e5d9bb6fc4ae5921657b7c00f95b91b30e921261900cc16a8ecc67","internal_anchors":2},"formal_canon":{"evidence_count":3,"snapshot_sha256":"8db3f21166dc6d078143d3af71924013ed4e8d639d48d53c8b4790320dfeba03"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2502.14802","created_at":"2026-05-17T23:38:45.966252+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.14802v2","created_at":"2026-05-17T23:38:45.966252+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.14802","created_at":"2026-05-17T23:38:45.966252+00:00"},{"alias_kind":"pith_short_12","alias_value":"T5TAHX3BRVAQ","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"T5TAHX3BRVAQGEPV","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"T5TAHX3B","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":23,"internal_anchor_count":23,"sample":[{"citing_arxiv_id":"2605.23753","citing_title":"SeedER: Seed-and-Expand Retrieval from Knowledge Graphs","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2603.23231","citing_title":"PERMA: Benchmarking Personalized Memory Agents via Event-Driven Preference and Realistic Task Environments","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2605.19366","citing_title":"Accurate, Efficient, and Explainable Deep Learning Approaches for Environmental Science Problems","ref_index":216,"is_internal_anchor":true},{"citing_arxiv_id":"2509.24276","citing_title":"G-reasoner: Foundation Models for Unified Reasoning over Graph-structured Knowledge","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2510.11541","citing_title":"Question-Adaptive Graph Learning for Multi-hop Retrieval Augmented Generation","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2504.15965","citing_title":"From Human Memory to AI Memory: A Survey on Memory Mechanisms in the Era of LLMs","ref_index":61,"is_internal_anchor":true},{"citing_arxiv_id":"2507.05257","citing_title":"Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2604.20844","citing_title":"AtomicRAG: Atom-Entity Graphs for Retrieval-Augmented Generation","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2603.23516","citing_title":"MSA: Memory Sparse Attention for Efficient End-to-End Memory Model Scaling to 100M Tokens","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2507.03724","citing_title":"MemOS: A Memory OS for AI System","ref_index":52,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13438","citing_title":"CogniFold: Always-On Proactive Memory via Cognitive Folding","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09666","citing_title":"Do We Still Need GraphRAG? Benchmarking RAG and GraphRAG for Agentic Search Systems","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2604.03384","citing_title":"BridgeRAG: Training-Free Bridge-Conditioned Retrieval for Multi-Hop Question Answering","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11814","citing_title":"MedMemoryBench: Benchmarking Agent Memory in Personalized Healthcare","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12061","citing_title":"SAGE: A Self-Evolving Agentic Graph-Memory Engine for Structure-Aware Associative Memory","ref_index":209,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09942","citing_title":"HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution","ref_index":48,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10488","citing_title":"DeepRefine: Agent-Compiled Knowledge Refinement via Reinforcement Learning","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00702","citing_title":"Learning How and What to Memorize: Cognition-Inspired Two-Stage Optimization for Evolving Memory","ref_index":126,"is_internal_anchor":true},{"citing_arxiv_id":"2604.11628","citing_title":"Back to Basics: Let Conversational Agents Remember with Just Retrieval and Generation","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09019","citing_title":"Regime-Conditional Retrieval: Theory and a Transferable Router for Two-Hop QA","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2604.06845","citing_title":"HingeMem: Boundary Guided Long-Term Memory with Query Adaptive Retrieval for Scalable Dialogues","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2604.14488","citing_title":"Controlling Authority Retrieval: A Missing Retrieval Objective for Authority-Governed Knowledge","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2604.17265","citing_title":"MemSearch-o1: Empowering Large Language Models with Reasoning-Aligned Memory Growth in Agentic Search","ref_index":18,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":3,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/T5TAHX3BRVAQGEPVQ32ZLBHQAA","json":"https://pith.science/pith/T5TAHX3BRVAQGEPVQ32ZLBHQAA.json","graph_json":"https://pith.science/api/pith-number/T5TAHX3BRVAQGEPVQ32ZLBHQAA/graph.json","events_json":"https://pith.science/api/pith-number/T5TAHX3BRVAQGEPVQ32ZLBHQAA/events.json","paper":"https://pith.science/paper/T5TAHX3B"},"agent_actions":{"view_html":"https://pith.science/pith/T5TAHX3BRVAQGEPVQ32ZLBHQAA","download_json":"https://pith.science/pith/T5TAHX3BRVAQGEPVQ32ZLBHQAA.json","view_paper":"https://pith.science/paper/T5TAHX3B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.14802&json=true","fetch_graph":"https://pith.science/api/pith-number/T5TAHX3BRVAQGEPVQ32ZLBHQAA/graph.json","fetch_events":"https://pith.science/api/pith-number/T5TAHX3BRVAQGEPVQ32ZLBHQAA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/T5TAHX3BRVAQGEPVQ32ZLBHQAA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/T5TAHX3BRVAQGEPVQ32ZLBHQAA/action/storage_attestation","attest_author":"https://pith.science/pith/T5TAHX3BRVAQGEPVQ32ZLBHQAA/action/author_attestation","sign_citation":"https://pith.science/pith/T5TAHX3BRVAQGEPVQ32ZLBHQAA/action/citation_signature","submit_replication":"https://pith.science/pith/T5TAHX3BRVAQGEPVQ32ZLBHQAA/action/replication_record"}},"created_at":"2026-05-17T23:38:45.966252+00:00","updated_at":"2026-05-17T23:38:45.966252+00:00"}