Get Started
Home
Topics
Search
Library
Research questionHow can a model select the correct entity-specific parametric memory for graph queries without query-time text or semantic similarity?Graph facts stored in separate parameter memories avoid repeating retrieved subgraphs in each prompt. However, a query’s answer may reside in one specific entity’s memory even when another entity is semantically similar, making ordinary similarity retrieval unreliable.
AI
AI Memory
Evaluation & Benchmarks
Information Retrieval
Machine Learning
Natural Language Processing
Retrieval-Augmented Generation
Latest papersRecent research connected to this question, newest first.A Storage-Retrieval Gap in Parametric Knowledge Graph MemoryEvidence comes from MetaQA using one LoRA adapter per entity, trained on subgraphs and tested on unseen questions. The findings cover single-valued relations, compare embedding- and weight-space geometry retrieval, and quantify storage and context-token costs against graph retrieval-augmented generation; they do not demonstrate a learned query-conditioned selector.research paper · Sep 7, 2026
Related questions
How can we construct accurate knowledge bases from an LLM’s parametric knowledge without retrieval, fine-tuning, or exceeding 32B parameters?How can robotic agents maintain queryable spatial-semantic memory for precise, low-latency answers in complex environments?How can graph processors preserve relational intermediate states when learning algorithmic computations?How can LLMs answer knowledge-intensive questions over large, unseen knowledge graphs without graph-specific tuning?