mlsanityData to 14 Oct 2026

AWM: Answerable Working Memory for Long-Document VQA Agents

arXiv 2608.25618 ↗First posted 26 Aug 2026cs.CLcited by 2

On the map: LLM reasoning → Multimodal reasoning

also in 1 more
Abstract (arXiv)

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 memory-only answerability, a diagnostic that asks whether a reader can answer from the question and terminal working memory alone. Building on this diagnostic, 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 textscMMLongBench-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 textscMMLongBench-Doc and textscLongDocURL and reduces the memory-missing-correct rate by 2.7 points over answer-only GRPO.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 36 of its 44 references that are in the map. First posted on arXiv 26 Aug 2026.

Builds on

36 of 44 references.

Closest ancestors · 3

Same direction, newest first

  1. 2026What's Missing in Screen-to-Action? Towards a UI-in-the-Loop Paradigm for Multimodal GUI Reasoning8 Apr
  2. 2025Perception-R1: Advancing Multimodal Reasoning Capabilities of MLLMs via Visual Perception Reward8 Jun
  3. 2024LongDocURL: a Comprehensive Multimodal Long Document Benchmark Integrating Understanding, Reasoning, and Locating24 Dec

Other references · 33

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  2. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  3. 2023Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflectioncited by 1,839
  4. 2023ChatGPT Outperforms Crowd-Workers for Text-Annotation Taskscited by 870
  5. 2023Enabling Large Language Models to Generate Text with Citationscited by 595
  6. 2023Evaluating Verifiability in Generative Search Enginescited by 303
  7. 2024VisRAG: Vision-based Retrieval-augmented Generation on Multi-modality Documentscited by 245
  8. 2024Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Modelscited by 235
  9. 2024MMLongBench-Doc: Benchmarking Long-context Document Understanding with Visualizationscited by 183
  10. 2024M3DocRAG: Multi-modal Retrieval is What You Need for Multi-page Multi-document Understandingcited by 111
  11. 2025VRAG-RL: Empower Vision-Perception-Based RAG for Visually Rich Information Understanding via Iterative Reasoning with Reinforcement Learningcited by 68
  12. 2025Unlocking Exploration in RLVR: Uncertainty-aware Advantage Shaping for Deeper Reasoningcited by 34
21 more
  1. 2025Visual Instance-aware Prompt Tuningcited by 29
  2. 2025MemSearcher: Training LLMs to Reason, Search and Manage Memory via End-to-End Reinforcement Learningcited by 27
  3. 2026Look Inward to Explore Outward: Learning Temperature Policy from LLM Internal States via Hierarchical RLcited by 27
  4. 2024SV-RAG: LoRA-Contextualizing Adaptation of MLLMs for Long Document Understandingcited by 25
  5. 2025Docopilot: Improving Multimodal Models for Document-Level Understandingcited by 19
  6. 2025MoLoRAG: Bootstrapping Document Understanding via Multi-modal Logic-aware Retrievalcited by 17
  7. 2025ArgRAG: Explainable Retrieval Augmented Generation using Quantitative Bipolar Argumentationcited by 12
  8. 2025When TableQA Meets Noise: A Dual Denoising Framework for Complex Questions and Large-scale Tablescited by 10
  9. 2024Predictive Multiplicity of Knowledge Graph Embeddings in Link Predictioncited by 9
  10. 2025What Breaks Knowledge Graph based RAG? Benchmarking and Empirical Insights into Reasoning under Incomplete Knowledgecited by 9
  11. 2025Auditing Data Membership in Reinforcement Learning With Verifiable Rewardscited by 8
  12. 2026Rethinking Table Pruning in TableQA: From Sequential Revisions to Gold Trajectory-Supervised Parallel Searchcited by 8
  13. 2025Evaluating Knowledge Graph Based Retrieval Augmented Generation Methods under Knowledge Incompletenesscited by 7
  14. 2025Enrich-on-Graph: Query-Graph Alignment for Complex Reasoning with LLM Enrichingcited by 7
  15. 2025Supposedly Equivalent Facts That Aren't? Entity Frequency in Pre-training Induces Asymmetry in LLMscited by 6
  16. 2026One Refiner to Unlock Them All: Inference-Time Reasoning Elicitation via Reinforcement Query Refinementcited by 6
  17. 2025URaG: Unified Retrieval and Generation in Multimodal LLMs for Efficient Long Document Understandingcited by 5
  18. 2026MM-Doc-R1: Training Agents for Long Document Visual Question Answering through Multi-turn Reinforcement Learningcited by 5
  19. 2025GR-Agent: Adaptive Graph Reasoning Agent under Incomplete Knowledgecited by 2
  20. 2026EDGE: Experience-Distillation for Guided Exploration in Agentic Reinforcement Learningcited by 2
  21. 2026EigentSearch-Q+: Enhancing Deep Research Agents with Structured Reasoning Toolscited by 1

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026MemoryWalker: Stop Training Agents on Contexts They Never Sawcited by 0
  2. 2026Uncertainty in Representation Learning on Knowledge Graphscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026InfoMem: Training Long-Context Memory Agents with Answer-Conditioned Information Gain
  2. 2026PMMC: Prospective Multimodal Memory Compilation for Long-Term LVLM Agents
  3. 2025Look Back to Reason Forward: Revisitable Memory for Long-Context LLM Agents
  4. 2026What Makes Agent Memory Useful for Reliable Unanswerable Question Handling?
  5. 2025MemR3: Memory Retrieval via Reflective Reasoning for LLM Agents
  6. 2026Ground Truth First: A Longitudinal Evaluation Instrument for Agent Memory, and the Tenure Crossover in Memory-Architecture Rankings
  7. 2026Mem++: Non-Destructive Memory for Long-Term Organizational LLM Agents
  8. 2026APDMem: Agent-Controlled Progressive Disclosure for Query-Adaptive Long-Term Memory