כתבה
arXiv cs.AI ·
Concept-Grounded Attention: תצורה מאובטחת של תשומת לב עם תצורה גרפית, גרסאות זמניות ומעמד ידע
Concept-Grounded Attention: A Controlled Evaluation of Graph-Injected Attention, Temporal Versioning, and Epistemic Status
מחקר חדש: Concept-Grounded Attention - תצורה מאובטחת של תשומת לב עם תצורה גרפית, גרסאות זמניות ומעמד ידע. המחקר כולל חידושים ב-LangGraph ו-Gemini.
תקציר מקורי באנגליתarXiv:2609.38684v1 Announce Type: new Abstract: Knowledge-intensive language-model systems typically represent external knowledge as text chunks or static graphs, with limited support for concept evolution, point-in-time reasoning, and distinctions between validated and inferred knowledge. We introduce the Concept Lifecycle Model (CLM), which represents concepts as persistent, graph-grounded, temporally versioned entities with explicit provenance and epistemic status, and Concept-Grounded Attention (CGA), which injects concept-graph structure into transformer computation through graph-biased self-attention (Form A) and gated cross-attention over concept nodes (Form B). We evaluate the framework in controlled settings using disabled-mechanism baselines. On 200 MuSiQue and HotpotQA questions
קרא במקור המקורי
arxiv.org
פתח כתבה מקורית