HAGE is a principled memory system for long-term conversation memory and multi-hop reasoning. It represents memory across four orthogonal relational graphs — Semantic, Temporal, Causal, and Entity — and introduces a co-evolutionary training framework that jointly optimizes trainable edge features and a query-conditioned QueryRouter MLP via policy-gradient reinforcement learning.</p>\n","updatedAt":"2026-05-14T03:58:23.143Z","author":{"_id":"680f20f5f3cd7c68f689e156","avatarUrl":"/avatars/b572737cbf6b14223770e497dc3ac895.svg","fullname":"dj","name":"dj220001","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.879371702671051},"editors":["dj220001"],"editorAvatarUrls":["/avatars/b572737cbf6b14223770e497dc3ac895.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.09942","authors":[{"_id":"6a0547eeb1a8cbabc9f0887a","name":"Dongming Jiang","hidden":false},{"_id":"6a0547eeb1a8cbabc9f0887b","name":"Yi Li","hidden":false},{"_id":"6a0547eeb1a8cbabc9f0887c","name":"Guanpeng Li","hidden":false},{"_id":"6a0547eeb1a8cbabc9f0887d","name":"Qiannan Li","hidden":false},{"_id":"6a0547eeb1a8cbabc9f0887e","name":"Bingzhe Li","hidden":false}],"publishedAt":"2026-05-11T00:00:00.000Z","submittedOnDailyAt":"2026-05-14T00:00:00.000Z","title":"HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution","submittedOnDailyBy":{"_id":"680f20f5f3cd7c68f689e156","avatarUrl":"/avatars/b572737cbf6b14223770e497dc3ac895.svg","isPro":false,"fullname":"dj","user":"dj220001","type":"user","name":"dj220001"},"summary":"Memory retrieval in agentic large language model (LLM) systems is often treated as a static lookup problem, relying on flat vector search or fixed binary relational graphs. However, fixed graph structures cannot capture the varying strength, confidence, and query-dependent relevance of relationships between events. In this paper, we propose HAGE, a weighted multi-relational memory framework that reconceptualizes retrieval as sequential, query-conditioned traversal over a unified relational memory graph. Memory is organized as relation-specific graph views over shared memory nodes, where each edge is associated with a trainable relation feature vector encoding multiple relational signals. Given a query, an LLM-based classifier identifies the relational intent, and a routing network dynamically modulates the corresponding dimensions of the edge embedding. Traversal scores are computed via a learned combination of semantic similarity and these query-conditioned edge representations. This allows memory traversal to prioritize high-utility relational paths while softly suppressing noisy or weakly relevant connections. Beyond adaptive traversal, HAGE further introduces a reinforcement learning-based training framework that jointly optimizes routing behavior and edge representations using downstream tasks. Finally, empirical results demonstrate improved long-horizon reasoning accuracy and a favorable accuracy-efficiency trade-off compared to state-of-the-art agentic memory systems. Our code is available at https://github.com/FredJiang0324/HAGE_MVPReview.","upvotes":10,"discussionId":"6a0547eeb1a8cbabc9f0887f","projectPage":"https://github.com/FredJiang0324/HAGE_MVPReview","githubRepo":"https://github.com/FredJiang0324/HAGE_MVPReview","githubRepoAddedBy":"user","ai_summary":"HAGE introduces a weighted multi-relational memory framework that enables query-conditioned traversal over unified relational memory graphs, improving long-horizon reasoning accuracy through adaptive memory retrieval and reinforcement learning-based optimization.","ai_keywords":["agentic large language model","memory retrieval","relational memory graph","relation-specific graph views","edge embedding","routing network","semantic similarity","reinforcement learning","downstream tasks","query-conditioned traversal"],"githubStars":11},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"680f20f5f3cd7c68f689e156","avatarUrl":"/avatars/b572737cbf6b14223770e497dc3ac895.svg","isPro":false,"fullname":"dj","user":"dj220001","type":"user"},{"_id":"6989005baa7d770d5185f500","avatarUrl":"/avatars/5453cfdc0324b459c4f88131963aa360.svg","isPro":false,"fullname":"Bingzhe Li","user":"libingzheren","type":"user"},{"_id":"66e9c93990465c96170d8630","avatarUrl":"/avatars/39f674083fcdf0367c60229c9ec730c8.svg","isPro":false,"fullname":"Songtao Wei","user":"Kotom1","type":"user"},{"_id":"6a0559a2f8a6e71075a68f83","avatarUrl":"/avatars/ee0590621428366a6d440db23648e975.svg","isPro":false,"fullname":"Dingyi Kang","user":"Dingy-Kang","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"69bb834398f36bb29439d57d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/X4UTVjFF3GJJqDpmO-IJD.png","isPro":false,"fullname":"林優奈","user":"smoore2024","type":"user"},{"_id":"69cca6345f1e1b0f7a07f72b","avatarUrl":"/avatars/70b92e65eae06470f79b5b77220731e5.svg","isPro":false,"fullname":"허주원","user":"jacksonhernande","type":"user"},{"_id":"64763dca71d420a1f6be634e","avatarUrl":"/avatars/e11dc1f3ff7c61883f16e8c04cc0871d.svg","isPro":false,"fullname":"PearLi","user":"PearMath","type":"user"},{"_id":"68308fb0cddb5ce16ae19445","avatarUrl":"/avatars/a5faa2cb600176a0050e06bb98193558.svg","isPro":false,"fullname":"Hao","user":"polar1ss","type":"user"},{"_id":"69ae7358ef5c1c3078061bf8","avatarUrl":"/avatars/cb2b3d881c72340cfd17813a3e2e0a19.svg","isPro":false,"fullname":"zhuangxian feng","user":"zhuangxian","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.09942.md"}">
HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
Published on May 11
· Submitted by dj on May 14 Abstract
HAGE introduces a weighted multi-relational memory framework that enables query-conditioned traversal over unified relational memory graphs, improving long-horizon reasoning accuracy through adaptive memory retrieval and reinforcement learning-based optimization.
AI-generated summary
Memory retrieval in agentic large language model (LLM) systems is often treated as a static lookup problem, relying on flat vector search or fixed binary relational graphs. However, fixed graph structures cannot capture the varying strength, confidence, and query-dependent relevance of relationships between events. In this paper, we propose HAGE, a weighted multi-relational memory framework that reconceptualizes retrieval as sequential, query-conditioned traversal over a unified relational memory graph. Memory is organized as relation-specific graph views over shared memory nodes, where each edge is associated with a trainable relation feature vector encoding multiple relational signals. Given a query, an LLM-based classifier identifies the relational intent, and a routing network dynamically modulates the corresponding dimensions of the edge embedding. Traversal scores are computed via a learned combination of semantic similarity and these query-conditioned edge representations. This allows memory traversal to prioritize high-utility relational paths while softly suppressing noisy or weakly relevant connections. Beyond adaptive traversal, HAGE further introduces a reinforcement learning-based training framework that jointly optimizes routing behavior and edge representations using downstream tasks. Finally, empirical results demonstrate improved long-horizon reasoning accuracy and a favorable accuracy-efficiency trade-off compared to state-of-the-art agentic memory systems. Our code is available at https://github.com/FredJiang0324/HAGE_MVPReview.
Community
HAGE is a principled memory system for long-term conversation memory and multi-hop reasoning. It represents memory across four orthogonal relational graphs — Semantic, Temporal, Causal, and Entity — and introduces a co-evolutionary training framework that jointly optimizes trainable edge features and a query-conditioned QueryRouter MLP via policy-gradient reinforcement learning.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2605.09942 in a model README.md to link it from this page.
Cite arxiv.org/abs/2605.09942 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2605.09942 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.