How Vision Becomes Language: A Layer-wise Information-Theoretic Analysis of Multimodal Reasoning

• URL: http://arxiv.org/abs/2602.15580v1
• Date: Tue, 17 Feb 2026 13:49:49 GMT
• Title: How Vision Becomes Language: A Layer-wise Information-Theoretic Analysis of Multimodal Reasoning
• Abstract summary: We show that visual-unique information peaks early and decays with depth, language-unique information surges in late layers, and cross-modal synergy remains below 2%.<n>These results provide an information-theoretic, causal account of how vision becomes language in multimodal Transformers.
• Score: 3.720343746691839
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: When a multimodal Transformer answers a visual question, is the prediction driven by visual evidence, linguistic reasoning, or genuinely fused cross-modal computation -- and how does this structure evolve across layers? We address this question with a layer-wise framework based on Partial Information Decomposition (PID) that decomposes the predictive information at each Transformer layer into redundant, vision-unique, language-unique, and synergistic components. To make PID tractable for high-dimensional neural representations, we introduce \emph{PID Flow}, a pipeline combining dimensionality reduction, normalizing-flow Gaussianization, and closed-form Gaussian PID estimation. Applying this framework to LLaVA-1.5-7B and LLaVA-1.6-7B across six GQA reasoning tasks, we uncover a consistent \emph{modal transduction} pattern: visual-unique information peaks early and decays with depth, language-unique information surges in late layers to account for roughly 82\% of the final prediction, and cross-modal synergy remains below 2\%. This trajectory is highly stable across model variants (layer-wise correlations $>$0.96) yet strongly task-dependent, with semantic redundancy governing the detailed information fingerprint. To establish causality, we perform targeted Image$\rightarrow$Question attention knockouts and show that disrupting the primary transduction pathway induces predictable increases in trapped visual-unique information, compensatory synergy, and total information cost -- effects that are strongest in vision-dependent tasks and weakest in high-redundancy tasks. Together, these results provide an information-theoretic, causal account of how vision becomes language in multimodal Transformers, and offer quantitative guidance for identifying architectural bottlenecks where modality-specific information is lost.

Related papers

• A Progressive Training Strategy for Vision-Language Models to Counteract Spatio-Temporal Hallucinations in Embodied Reasoning [49.61652671596548]
  "Multi-image hallucination reasoning", where a massive performance drop between forward and temporal queries reveals a dependence on superficial shortcuts instead of genuine understanding.<n>To mitigate this, we develop a new dataset: Chain-of-temporal building on detailed reasoning into Chain steps and definitive judgments.<n>Our experiments demonstrate that this approach not only improves accuracy but also the forward-backward performance gap from over 70% to only 6.53%.
  arXiv  Detail & Related papers  (2026-04-12T07:48:44Z)
• Insight-V++: Towards Advanced Long-Chain Visual Reasoning with Multimodal Large Language Models [65.4947731385794]
  We present a unified multi-agent visual reasoning framework that evolves from our foundational image-centric model, Insight-V.<n>We introduce two novel algorithms, ST-GRPO and J-GRPO, which enhance spatial-temporal reasoning and improve evaluative robustness.<n>Experiments on base models like LLaVA-NeXT and Qwen2.5-VL demonstrate significant performance gains across challenging image and video reasoning benchmarks.
  arXiv  Detail & Related papers  (2026-03-18T15:28:07Z)
• Mechanisms of Non-Monotonic Scaling in Vision Transformers [0.0]
  We identify a consistent three-phase Cliff-au-Climb pattern that governs how representations evolve with depth.<n>We quantify patterns of information mixing with an Information Scrambling Index, and show that in ViT-L the information-task tradeoff emerges roughly 10 layers later than in ViT-B.
  arXiv  Detail & Related papers  (2025-11-26T18:07:14Z)
• Identifying Super Spreaders in Multilayer Networks [0.6990493129893112]
  We introduce a novel approach to identifying super-spreaders in such networks by leveraging graph neural networks.<n>To this end, we construct a dataset by simulating information diffusion across hundreds of networks.<n>Our model, TopSpreadersNetwork, comprises a relationship-agnostic encoder and a custom aggregation layer.
  arXiv  Detail & Related papers  (2025-05-27T10:14:14Z)
• Transformer Meets Twicing: Harnessing Unattended Residual Information [2.1605931466490795]
  Transformer-based deep learning models have achieved state-of-the-art performance across numerous language and vision tasks.<n>While the self-attention mechanism has proven capable of handling complex data patterns, it has been observed that the representational capacity of the attention matrix degrades significantly across transformer layers.<n>We propose the Twicing Attention, a novel attention mechanism that uses kernel twicing procedure in nonparametric regression to alleviate the low-pass behavior of associated NLM smoothing.
  arXiv  Detail & Related papers  (2025-03-02T01:56:35Z)
• What Matters When Repurposing Diffusion Models for General Dense Perception Tasks? [49.84679952948808]
  Recent works show promising results by simply fine-tuning T2I diffusion models for dense perception tasks.<n>We conduct a thorough investigation into critical factors that affect transfer efficiency and performance when using diffusion priors.<n>Our work culminates in the development of GenPercept, an effective deterministic one-step fine-tuning paradigm tailed for dense visual perception tasks.
  arXiv  Detail & Related papers  (2024-03-10T04:23:24Z)
• Disentangled Representation Learning with Transmitted Information Bottleneck [57.22757813140418]
  We present textbfDisTIB (textbfTransmitted textbfInformation textbfBottleneck for textbfDisd representation learning), a novel objective that navigates the balance between information compression and preservation.
  arXiv  Detail & Related papers  (2023-11-03T03:18:40Z)
• Frozen Overparameterization: A Double Descent Perspective on Transfer Learning of Deep Neural Networks [27.17697714584768]
  We study the generalization behavior of transfer learning of deep neural networks (DNNs) We show that the test error evolution during the target training has a more significant double descent effect when the target training dataset is sufficiently large. Also, we show that the double descent phenomenon may make a transfer from a less related source task better than a transfer from a more related source task.
  arXiv  Detail & Related papers  (2022-11-20T20:26:23Z)
• A Generic Shared Attention Mechanism for Various Backbone Neural Networks [53.36677373145012]
  Self-attention modules (SAMs) produce strongly correlated attention maps across different layers. Dense-and-Implicit Attention (DIA) shares SAMs across layers and employs a long short-term memory module. Our simple yet effective DIA can consistently enhance various network backbones.
  arXiv  Detail & Related papers  (2022-10-27T13:24:08Z)
• Correlation Information Bottleneck: Towards Adapting Pretrained Multimodal Models for Robust Visual Question Answering [63.87200781247364]
  Correlation Information Bottleneck (CIB) seeks a tradeoff between compression and redundancy in representations. We derive a tight theoretical upper bound for the mutual information between multimodal inputs and representations.
  arXiv  Detail & Related papers  (2022-09-14T22:04:10Z)
• DepthFormer: Exploiting Long-Range Correlation and Local Information for Accurate Monocular Depth Estimation [50.08080424613603]
  Long-range correlation is essential for accurate monocular depth estimation. We propose to leverage the Transformer to model this global context with an effective attention mechanism. Our proposed model, termed DepthFormer, surpasses state-of-the-art monocular depth estimation methods with prominent margins.
  arXiv  Detail & Related papers  (2022-03-27T05:03:56Z)
• Handling Distribution Shifts on Graphs: An Invariance Perspective [78.31180235269035]
  We formulate the OOD problem on graphs and develop a new invariant learning approach, Explore-to-Extrapolate Risk Minimization (EERM) EERM resorts to multiple context explorers that are adversarially trained to maximize the variance of risks from multiple virtual environments. We prove the validity of our method by theoretically showing its guarantee of a valid OOD solution.
  arXiv  Detail & Related papers  (2022-02-05T02:31:01Z)
• Understanding Self-supervised Learning with Dual Deep Networks [74.92916579635336]
  We propose a novel framework to understand contrastive self-supervised learning (SSL) methods that employ dual pairs of deep ReLU networks. We prove that in each SGD update of SimCLR with various loss functions, the weights at each layer are updated by a emphcovariance operator. To further study what role the covariance operator plays and which features are learned in such a process, we model data generation and augmentation processes through a emphhierarchical latent tree model (HLTM)
  arXiv  Detail & Related papers  (2020-10-01T17:51:49Z)

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.