Responses Fall Short of Understanding: Revealing the Gap between Internal Representations and Responses in Visual Document Understanding

• URL: http://arxiv.org/abs/2604.04411v1
• Date: Mon, 06 Apr 2026 04:25:52 GMT
• Title: Responses Fall Short of Understanding: Revealing the Gap between Internal Representations and Responses in Visual Document Understanding
• Abstract summary: We investigate how information required to solve VDU tasks is represented across different layers of large vision language models.<n>Our study reveals that there is a clear gap between internal representations and generated responses.<n>Experiments show that fine-tuning intermediate layers improves both linear probing accuracy and response accuracy while narrowing the gap.
• Score: 13.93076527078901
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Visual document understanding (VDU) is a challenging task for large vision language models (LVLMs), requiring the integration of visual perception, text recognition, and reasoning over structured layouts. Although recent LVLMs have shown progress on VDU benchmarks, their performance is typically evaluated based on generated responses, which may not necessarily reflect whether the model has actually captured the required information internally. In this paper, we investigate how information required to solve VDU tasks is represented across different layers of LLMs within LVLMs using linear probing. Our study reveals that (1) there is a clear gap between internal representations and generated responses, and (2) information required to solve the task is often encoded more linearly from intermediate layers than from the final layer. Motivated by these findings, we explore fine-tuning strategies that target intermediate layers. Experiments show that fine-tuning intermediate layers improves both linear probing accuracy and response accuracy while narrowing the gap.

Related papers

• Diagnosing Visual Ignorance in Vision-Language Models [29.2851901986069]
  Vision-Language Models (VLMs) frequently rely on language priors, producing confident answers that are weakly grounded in visual evidence.<n>In this work, we study language-prior reliance from both mechanistic and behavioral perspectives.<n>We find that language-prior reliance is a systematic routing failure affecting both model internals and benchmark validity.
  arXiv  Detail & Related papers  (2026-06-05T04:16:01Z)
• Predictive Regularization Against Visual Representation Degradation in Multimodal Large Language Models [84.94288033791346]
  We conduct a detailed diagnostic analysis to unveil a pervasive issue: visual representation degradation in MLLMs.<n>We attribute this phenomenon to a visual sacrifice driven by the singular text-generation objective, where the model compromises its visual fidelity to optimize for answer generation.<n>We propose Predictive Regularization to force degraded intermediate features to predict initial visual features, thereby maintaining the inherent visual attributes of the MLLM's internal representations.
  arXiv  Detail & Related papers  (2026-03-21T13:10:37Z)
• From Drop-off to Recovery: A Mechanistic Analysis of Segmentation in MLLMs [50.185593677108436]
  Multimodal Large Language Models (MLLMs) are increasingly applied to pixel-level vision tasks, yet their intrinsic capacity for spatial understanding remains poorly understood.<n>We investigate segmentation capacity through a layerwise linear probing evaluation across the entire MLLM pipeline: vision encoder, adapter, and LLM.
  arXiv  Detail & Related papers  (2026-03-18T00:22:15Z)
• A Comprehensive Study on Visual Token Redundancy for Discrete Diffusion-based Multimodal Large Language Models [85.30893355216486]
  We study how visual token redundancy evolves with different dMLLM architectures and tasks.<n>Our study reveals that visual redundancy emerges only in from-scratch dMLLMs while handling long-answer tasks.<n>Layer-skipping is promising for accelerating AR-to-diffusion dMLLMs, whereas progressive or late-step pruning is more effective for from-scratch dMLLMs.
  arXiv  Detail & Related papers  (2025-11-19T04:13:36Z)
• How Multimodal LLMs Solve Image Tasks: A Lens on Visual Grounding, Task Reasoning, and Answer Decoding [39.342366994703376]
  We introduce a probing framework to analyze how MLLMs process visual and textual inputs across layers.<n>We show that while the overall stage-wise structure remains stable across variations in visual tokenization, instruction tuning data, and pretraining corpus, the specific layer allocation to each stage shifts.
  arXiv  Detail & Related papers  (2025-08-27T21:22:01Z)
• ONLY: One-Layer Intervention Sufficiently Mitigates Hallucinations in Large Vision-Language Models [67.75439511654078]
  Large Vision-Language Models (LVLMs) have introduced a new paradigm for understanding and reasoning about image input through textual responses.<n>They face the persistent challenge of hallucination, which introduces practical weaknesses and raises concerns about their reliable deployment in real-world applications.<n>We propose ONLY, a training-free decoding approach that requires only a single query and a one-layer intervention during decoding, enabling efficient real-time deployment.
  arXiv  Detail & Related papers  (2025-07-01T16:01:08Z)
• mRAG: Elucidating the Design Space of Multi-modal Retrieval-Augmented Generation [15.991125806837386]
  Large Vision-Language Models (LVLMs) have made remarkable strides in multimodal tasks such as visual question answering, visual grounding, and complex reasoning.<n>Retrieval-Augmented Generation (RAG) offers a practical solution to mitigate these challenges by allowing the LVLMs to access large-scale knowledge databases via retrieval mechanisms.
  arXiv  Detail & Related papers  (2025-05-29T23:32:03Z)
• Are Unified Vision-Language Models Necessary: Generalization Across Understanding and Generation [50.22361866757033]
  unified vision-language models (VLMs) integrate both visual understanding and generation capabilities.<n>This paper systematically investigates the generalization across understanding and generation tasks in unifiedVLMs.
  arXiv  Detail & Related papers  (2025-05-29T03:40:21Z)
• Multimodal Language Models See Better When They Look Shallower [54.5303326937134]
  Multimodal large language models (MLLMs) typically extract visual features from the final layers of a pretrained Vision Transformer (ViT)<n>We present the first comprehensive study of visual layer selection for MLLMs, analyzing representation similarity across ViT layers.<n>We find that while deep layers excel in semantic-rich tasks like OCR, shallow and middle layers significantly outperform them on fine-grained visual tasks.
  arXiv  Detail & Related papers  (2025-04-30T09:07:10Z)
• Does Representation Matter? Exploring Intermediate Layers in Large Language Models [22.704926222438456]
  We investigate the quality of intermediate representations in large language models (LLMs)<n>We find that intermediate layers often yield more informative representations for downstream tasks than the final layers.<n>Our results illuminate the internal mechanics of LLMs and guide strategies for architectural optimization and training.
  arXiv  Detail & Related papers  (2024-12-12T18:48:51Z)
• From Redundancy to Relevance: Information Flow in LVLMs Across Reasoning Tasks [33.476693301050275]
  We conduct experiments with truncation strategies across various LVLMs for visual question answering and image captioning tasks. By exploring the information flow from the perspective of visual representation contribution, we observe that it tends to converge in shallow layers but diversify in deeper layers.
  arXiv  Detail & Related papers  (2024-06-04T13:52:54Z)
• 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)

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.