Layers, Sinks, and Scaling: Adaptive Evidence Selection for Multimodal Large Language Models

• URL: http://arxiv.org/abs/2609.16795v1
• Date: Tue, 15 Sep 2026 08:02:52 GMT
• Title: Layers, Sinks, and Scaling: Adaptive Evidence Selection for Multimodal Large Language Models
• Abstract summary: Multimodal large language models (MLLMs) can answer knowledge-intensive visual questions by combining visual evidence from images with facts retrieved from external sources.<n>MLLMs may overlook relevant evidence in both modalities, attending weakly to the textual sentences or visual regions needed for the correct answer.<n>Recent efforts address this by highlighting retrieved text and marking visual regions before generation, but apply a fixed, one-shot policy.<n>We introduce Adaptive Relevance-guided Evidence Allocation (AREA), a training-free inference-time method that formulates evidence highlighting as adaptive allocation.
• Score: 17.809800994999225
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multimodal large language models (MLLMs) can answer knowledge-intensive visual questions by combining visual evidence from images with facts retrieved from external sources. However, MLLMs may overlook relevant evidence in both modalities, attending weakly to the textual sentences or visual regions needed for the correct answer. Recent efforts address this by highlighting retrieved text and marking visual regions before generation, but apply a fixed, one-shot policy that cannot adapt to three sources of variation: whether highlighting is necessary, how much evidence different examples require, and when different textual evidence becomes relevant as the answer unfolds. We introduce Adaptive Relevance-guided Evidence Allocation (AREA), a training-free inference-time method that formulates evidence highlighting as adaptive allocation. AREA generates a single probe token to read visual and textual relevance from fixed backbone layers, then makes three decisions: i) whether to intervene (controlled by natural attention coverage and visual sink contamination), ii) how much evidence to expose (determined by relevance entropy), and iii) when to refresh text during generation (triggered by causal context-attention peaks). Across four KB-VQA and seven standard multimodal benchmarks with nine frozen MLLM checkpoints, establishes the best performance among training-free highlighting methods.

Related papers

• Look Twice: Training-Free Evidence Highlighting in Multimodal Large Language Models [27.14308995992974]
  We introduce Look Twice (LoT), a training-free inference-time framework that improves how pretrained MLLMs utilize multimodal evidence.<n>LoT estimates which visual regions and retrieved textual elements are relevant to a query, and then generates the answer conditioned on this highlighted evidence.<n> Experiments across multiple knowledge-based VQA benchmarks show consistent improvements over zero-shot MLLMs.
  arXiv  Detail & Related papers  (2026-04-01T18:00:08Z)
• MARA: A Multimodal Adaptive Retrieval-Augmented Framework for Document Question Answering [51.19392014547221]
  Retrieval-based multimodal document QA aims to identify and integrate relevant information from visually rich documents with complex multimodal structures.<n>Current approaches rely on query-agnostic document representations that overlook salient content.<n>We propose the Multimodal Adaptive Retrieval-Augmented (MARA) framework, which introduces query-adaptive generation.
  arXiv  Detail & Related papers  (2026-02-01T12:27:40Z)
• Do Images Speak Louder than Words? Investigating the Effect of Textual Misinformation in VLMs [17.56537230934894]
  Vision-Language Models (VLMs) have shown strong multimodal reasoning capabilities on Visual-Question-Answering (VQA) benchmarks.<n>We show that these models are indeed vulnerable to misleading textual prompts, often overriding clear visual evidence in favor of the conflicting text.
  arXiv  Detail & Related papers  (2026-01-27T05:04:38Z)
• DIVER: Dynamic Iterative Visual Evidence Reasoning for Multimodal Fake News Detection [6.225860651499494]
  Multimodal fake news detection is crucial for mitigating adversarial misinformation.<n>We propose DIVER (Dynamic Iterative Visual Evidence Reasoning), a framework grounded in a progressive, evidence-driven reasoning paradigm.<n>Experiments on Weibo, Weibo21, and GossipCop demonstrate that DIVER outperforms state-of-the-art baselines by an average of 2.72%.
  arXiv  Detail & Related papers  (2026-01-12T04:01:33Z)
• ReAG: Reasoning-Augmented Generation for Knowledge-based Visual Question Answering [54.72902502486611]
  ReAG is a Reasoning-Augmented Multimodal RAG approach that combines coarse- and fine-grained retrieval with a critic model that filters irrelevant passages.<n>ReAG significantly outperforms prior methods, improving answer accuracy and providing interpretable reasoning grounded in retrieved evidence.
  arXiv  Detail & Related papers  (2025-11-27T19:01:02Z)
• MAVIS: A Benchmark for Multimodal Source Attribution in Long-form Visual Question Answering [44.41273615523289]
  We introduce MAVIS, the first benchmark designed to evaluate multimodal source attribution systems.<n>Our dataset comprises 157K visual QA instances, where each answer is annotated with fact-level citations referring to multimodal documents.<n>We develop fine-grained automatic metrics along three dimensions of informativeness, groundedness, and fluency, and demonstrate their strong correlation with human judgments.
  arXiv  Detail & Related papers  (2025-11-15T10:14:59Z)
• Benchmarking Retrieval-Augmented Multimodal Generation for Document Question Answering [60.062194349648195]
  Document Visual Question Answering (DocVQA) faces dual challenges in processing lengthy multimodal documents.<n>Current document retrieval-augmented generation (DocRAG) methods remain limited by their text-centric approaches.<n>We introduce MMDocRAG, a comprehensive benchmark featuring 4,055 expert-annotated QA pairs with multi-page, cross-modal evidence chains.
  arXiv  Detail & Related papers  (2025-05-22T09:52:57Z)
• VLM-R$^3$: Region Recognition, Reasoning, and Refinement for Enhanced Multimodal Chain-of-Thought [51.43082554363725]
  We introduce textbfVLM-R$3$ (textbfVisual textbfLanguage textbfModel with textbfRegion textbfRecognition and textbfReasoning), a framework that equips an MLLM with the ability to decide emph when additional visual evidence is needed.<n>Experiments on MathVista, ScienceQA, and other benchmarks show that VLM-R$3$ sets a new
  arXiv  Detail & Related papers  (2025-05-22T03:50:13Z)
• MC-Bench: A Benchmark for Multi-Context Visual Grounding in the Era of MLLMs [61.56904387052982]
  This paper proposes a new visual grounding task called multi-context visual grounding.<n>It aims to localize instances of interest across multiple images based on open-ended text prompts.<n>We benchmark over 20 state-of-the-art MLLMs and foundation models with potential multi-context visual grounding capabilities.
  arXiv  Detail & Related papers  (2024-10-16T07:52:57Z)
• Multimodal Misinformation Detection using Large Vision-Language Models [7.505532091249881]
  Large language models (LLMs) have shown remarkable performance in various tasks. Few approaches consider evidence retrieval as part of misinformation detection. We propose a novel re-ranking approach for multimodal evidence retrieval.
  arXiv  Detail & Related papers  (2024-07-19T13:57:11Z)
• Debiasing Multimodal Large Language Models via Penalization of Language Priors [38.97645845493758]
  Multimodal Large Language Models (MLLMs) have become indispensable tools in computer vision and natural language processing.<n>Despite their advancements, our investigation reveals a noteworthy bias: the generated content is often driven more by the inherent priors of the underlying Large Language Models (LLMs) than by the input image.<n>We propose two simple, training-free strategies to rectify these biases and redirect the model's focus toward visual information.
  arXiv  Detail & Related papers  (2024-03-08T12:35:07Z)
• Contextual Object Detection with Multimodal Large Language Models [66.15566719178327]
  We introduce a novel research problem of contextual object detection. Three representative scenarios are investigated, including the language cloze test, visual captioning, and question answering. We present ContextDET, a unified multimodal model that is capable of end-to-end differentiable modeling of visual-language contexts.
  arXiv  Detail & Related papers  (2023-05-29T17:50:33Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.