AWM: Answerable Working Memory for Long-Document VQA Agents

2026-08-26

AI Summary

研究解决了长文档VQA中工作记忆质量低的问题，通过引入可回答的工作记忆(AWM)方法，改善了最终答案准确性和记忆的完整性。

Abstract

Long-document visual question answering increasingly relies on VLM agents that retrieve candidate pages, inspect page images, write findings to working memory, and synthesize answers. Working memory should carry answer-supporting evidence across page inspections for later grounded answering, yet existing evaluation mainly checks final-answer correctness and evidence-page access. This creates a memory-quality blind spot: an agent may reach the right page and answer correctly while leaving behind memory too generic or incomplete to support answering once page context is removed. We introduce \emph{memory-only answerability}, a diagnostic that asks whether a reader can answer from the question and terminal working memory alone. Building on this diagnostic, \emph{Answerable Working Memory} (AWM) treats terminal working memory as an answerable evidence artifact, and AWM-GRPO incorporates this signal into the GRPO reward while preserving final-answer priority. Under GRPO, this reward assigns higher advantages to answer-correct trajectories whose terminal working memory remains answerable. On \textsc{MMLongBench-Doc}, even when gold evidence pages are provided, 42.5\% of correct answers still cannot be answered from terminal working memory alone. AWM-GRPO improves final-answer accuracy over the RAG baseline by 8.1 and 11.9 points on \textsc{MMLongBench-Doc} and \textsc{LongDocURL} and reduces the memory-missing-correct rate by 2.7 points over answer-only GRPO.

Problem

• Long-Document VQA
• Working Memory
• Memory Quality
• Evidence Artifact

Innovation

• memory-only answerability
• Answerable Working Memory (AWM)
• AWM-GRPO