Fill the GAP: A Granular Alignment Paradigm for Visual Reasoning in Multimodal Large Language Models

• URL: http://arxiv.org/abs/2605.12374v3
• Date: Wed, 20 May 2026 07:37:13 GMT
• Title: Fill the GAP: A Granular Alignment Paradigm for Visual Reasoning in Multimodal Large Language Models
• Abstract summary: Visual latent reasoning lets a multimodal large language model (MLLM) create intermediate visual evidence as continuous tokens.<n>We propose textbfGAP, a textbfGranular textbfAlignment textbfParadigm for visual latent modeling.<n>On Qwen2.5-VL 7B, the resulting model achieves the best mean aggregate perception and reasoning performance among our supervised variants.
• Score: 35.34635490741956
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Visual latent reasoning lets a multimodal large language model (MLLM) create intermediate visual evidence as continuous tokens, avoiding external tools or image generators. However, existing methods usually follow an output-as-input latent paradigm and yield unstable gains. We identify evidence for a feature-space mismatch that can contribute to this instability: dominant visual-latent models build on pre-norm MLLMs and reuse decoder hidden states as predicted latent inputs, even though these states occupy a substantially different norm regime from the input embeddings the model was trained to consume~\citep{xie2025mhc,li2026siamesenorm,team2026attention}. This mismatch can make direct latent feedback unreliable. Motivated by this diagnosis, we propose \textbf{GAP}, a \textbf{G}ranular \textbf{A}lignment \textbf{P}aradigm for visual latent modeling. GAP aligns visual latent reasoning at three levels: feature-level alignment maps decoder outputs into input-compatible visual latents through a lightweight PCA-aligned latent head; context-level alignment grounds latent targets with inspectable auxiliary visual supervision; and capacity-guided alignment assigns latent supervision selectively to examples where the base MLLM struggles. On Qwen2.5-VL 7B, the resulting model achieves the best mean aggregate perception and reasoning performance among our supervised variants. Inference-time intervention probing further suggests that generated latents provide task-relevant visual signal beyond merely adding token slots.

Related papers

• Latent-Aligned Reasoning for Multimodal Recommendation [4.868819130885598]
  We propose LARK, a two-stage latent reasoning framework with complementary alignment mechanisms within a single VLM.<n>In the first stage, learnable latent tokens are interleaved with multi-step chain-of-thought reasoning.<n>In the second stage, latent representations are projected via a bridge and trained with item-to-item contrastive learning.
  arXiv  Detail & Related papers  (2026-09-04T02:24:48Z)
• Steering Vision-Language Models with Joint Sparse Autoencoders [2.367986397868328]
  We introduce the Joint Sparse Autoencoder (JSAE) to factorize sequence-pooled vision and language activations into shared, interpretable image/caption-level features.<n>Applying JSAE to vision-language models (VLMs) yields representations that are difficult to use as controllable cross-modal steering directions.
  arXiv  Detail & Related papers  (2026-06-24T10:11:04Z)
• Self-Prophetic Decoding to Unlock Visual Search in LVLMs [108.77389957341586]
  Large Vision-Language Models (LVLMs) are rapidly evolving toward true multimodal reasoning.<n>LVLM visual search faces two key challenges: incompatibility among intrinsic capabilities after post-training, and interference in long multi-step reasoning contexts.<n>We introduce SeProD, a self-prophetic decoding framework that leverages intrinsic single-step capabilities to enable coherent multi-step reasoning.
  arXiv  Detail & Related papers  (2026-05-27T17:01:39Z)
• Enhancing Multimodal In-Context Learning via Inductive-Deductive Reasoning [65.15766304205657]
  In-context learning (ICL) allows large models to adapt to tasks using a few examples, yet its extension to vision-language models (VLMs) remains fragile.<n>Our analysis reveals that the fundamental limitation lies in an inductive gap, models often produce correct answers from flawed reasoning.<n>We introduce a framework that restructures multimodal ICL as a principled inductive-deductive process.
  arXiv  Detail & Related papers  (2026-05-04T09:18:19Z)
• Online Self-Calibration Against Hallucination in Vision-Language Models [23.13137973421435]
  Large Vision-Language Models (LVLMs) often suffer from hallucinations, generating descriptions that include visual details absent from the input image.<n>We propose textbfOnline textbfSelf-textbfCAlibtextbfRation (OSCAR), a framework that integrates Monte Carlo Tree Search with a Dual-Granularity Reward Mechanism to construct preference data.
  arXiv  Detail & Related papers  (2026-05-01T01:03:05Z)
• Not All Tokens See Equally: Perception-Grounded Policy Optimization for Large Vision-Language Models [38.47027398567909]
  Perception-Grounded Policy Optimization (PGPO) is a novel fine-grained credit assignment framework that dynamically reshapes advantages at the token level.<n>We show that PGPO actively amplifies learning signals for visually-dependent tokens while suppressing gradient noise from linguistic priors.<n>Both theoretical and empirical analyses confirm that PGPO effectively reduces gradient variance, prevents training collapse, and acts as a potent regularizer for robust, perception-grounded multimodal reasoning.
  arXiv  Detail & Related papers  (2026-04-02T09:53:20Z)
• Sparse Visual Thought Circuits in Vision-Language Models [2.5754366051855837]
  We develop a causal pipeline to localize and test sparse visual thought circuits in Qwen3-VL-8B.<n>We train SAEs at this layer, construct task-selective sets via an explicit rule, and perform inference time scaling and ablation while quantifying accuracy and drift.
  arXiv  Detail & Related papers  (2026-03-26T06:24:36Z)
• Perception-Aware Multimodal Spatial Reasoning from Monocular Images [57.42071289037214]
  spatial reasoning from monocular images is essential for autonomous driving.<n>Current Vision-Language Models (VLMs) still struggle with fine-grained geometric perception.<n>We propose a perception-aware multimodal reasoning framework that equipsVLMs with explicit object-centric grounding ability.
  arXiv  Detail & Related papers  (2026-03-07T02:05:12Z)
• CrystaL: Spontaneous Emergence of Visual Latents in MLLMs [55.34169914483764]
  We propose CrystaL (Crystallized Latent Reasoning), a single-stage framework with two paths to process intact and corrupted images.<n>By explicitly aligning the attention patterns and prediction distributions across the two paths, CrystaL crystallizes latent representations into task-relevant visual semantics.<n>Experiments on perception-intensive benchmarks demonstrate that CrystaL consistently outperforms state-of-the-art baselines.
  arXiv  Detail & Related papers  (2026-02-24T15:01:30Z)
• SGHA-Attack: Semantic-Guided Hierarchical Alignment for Transferable Targeted Attacks on Vision-Language Models [73.19044613922911]
  Large vision-language models (VLMs) are vulnerable to transfer-based adversarial perturbations.<n>We propose SGHA-Attack, a framework that adopts multiple target references and enforces intermediate-layer consistency.<n>Experiments on open-source and commercial black-box VLMs show that SGHA-Attack achieves stronger targeted transferability than prior methods.
  arXiv  Detail & Related papers  (2026-02-02T03:10:41Z)
• FRISM: Fine-Grained Reasoning Injection via Subspace-Level Model Merging for Vision-Language Models [20.47311573790516]
  We propose FRISM (Fine-grained Reasoning Injection via Subspace-level model Merging), a fine-grained reasoning injection framework based on subspace-level model merging.<n>Experiments demonstrate that FRISM effectively improves reasoning capabilities without compromising the model's original visual capabilities.
  arXiv  Detail & Related papers  (2026-01-29T02:36:19Z)
• Mitigating Multimodal Hallucinations via Gradient-based Self-Reflection [49.26064449816502]
  We propose a Gradient-based Influence-Aware Constrained Decoding (GACD) method to address text-visual bias and co-occurrence bias.<n>GACD effectively reduces hallucinations and improves the visual grounding of MLLM outputs.
  arXiv  Detail & Related papers  (2025-09-03T08:13:52Z)
• GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs [56.93583799109029]
  GrAInS is an inference-time steering approach that operates across both language-only and vision-language models and tasks.<n>During inference, GrAInS hidden activations at transformer layers guided by token-level attribution signals, and normalizes activations to preserve representational scale.<n>It consistently outperforms both fine-tuning and existing steering baselines.
  arXiv  Detail & Related papers  (2025-07-24T02:34:13Z)

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.