Efficient Table QA via TableGrid Navigation and Progressive Inference Prompting

This arXiv feed (May 21, 2026) collects CS/IR papers about LLM-driven retrieval, long-term memory, efficiency, and domain-specific deployments. Highlights include: TableGrid Navigation (TGN) and Progressive Inference Prompting (PIP) for more precise table QA; Layer-wise Token Compression (LTC) to speed cross-encoder rerankers; CALMem, an application-layer dual-memory architecture that gives LLM assistants effectively unbounded context without model changes; MemConflict, a diagnostic benchmark showing how long-term memory systems fail under temporal/factual/contextual conflicts; GraphRAG-on-CPU