CrystaL: Spontaneous Emergence of Visual Latents in MLLMs

• URL: http://arxiv.org/abs/2602.20980v1
• Date: Tue, 24 Feb 2026 15:01:30 GMT
• Title: CrystaL: Spontaneous Emergence of Visual Latents in MLLMs
• Authors: Yang Zhang, Danyang Li, Yuxuan Li, Xin Zhang, Tianyu Xie, Mingming Cheng, Xiang Li,
• Abstract summary: 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.
• Score: 55.34169914483764
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multimodal Large Language Models (MLLMs) have achieved remarkable performance by integrating powerful language backbones with large-scale visual encoders. Among these, latent Chain-of-Thought (CoT) methods enable implicit reasoning in continuous hidden states, facilitating seamless vision-language integration and faster inference. However, existing heuristically predefined supervision signals in latent CoT provide limited guidance for preserving critical visual information in intermediate latent states. To address this limitation, we propose CrystaL (Crystallized Latent Reasoning), a single-stage framework with two paths to process intact and corrupted images, respectively. By explicitly aligning the attention patterns and prediction distributions across the two paths, CrystaL crystallizes latent representations into task-relevant visual semantics, without relying on auxiliary annotations or external modules. Extensive experiments on perception-intensive benchmarks demonstrate that CrystaL consistently outperforms state-of-the-art baselines, achieving substantial gains in fine-grained visual understanding while maintaining robust reasoning capabilities.

Related papers

• Do MLLMs Really See It: Reinforcing Visual Attention in Multimodal LLMs [55.61018839017648]
  Chain-of-thought (CoT) reasoning has substantially improved multimodal large language models (MLLMs) on complex reasoning tasks.<n>Existing approaches largely rely on long textual reasoning trajectories and provide limited mechanisms for learning stable visual attention policies.<n>We propose SAYO, a visual reasoning model trained with a reinforcement learning framework that introduces a region-level visual attention-based reward.
  arXiv  Detail & Related papers  (2026-02-09T03:33:23Z)
• ViTCoP: Accelerating Large Vision-Language Models via Visual and Textual Semantic Collaborative Pruning [8.933549837045932]
  Large Vision-Language Models incur high computational costs due to significant redundancy in their visual tokens.<n>We propose a Visual and Textual Semantic Collaborative Pruning framework (ViTCoP) that combines redundancy filtering in the vision encoder with step-wise co-pruning within the Large Language Models.
  arXiv  Detail & Related papers  (2026-01-25T12:47:30Z)
• Stable Language Guidance for Vision-Language-Action Models [62.80963701282789]
  Residual Semantic Steering is a probabilistic framework that disentangles physical affordance from semantic execution.<n> RSS achieves state-of-the-art robustness, maintaining performance even under adversarial linguistic perturbations.
  arXiv  Detail & Related papers  (2026-01-07T16:16:10Z)
• Monet: Reasoning in Latent Visual Space Beyond Images and Language [55.424507246294326]
  "Thinking with images" has emerged as an effective paradigm for advancing visual reasoning.<n>Existing methods fall short of human-like abstract visual thinking.<n>We introduce Monet, a training framework that enables multimodal large language models to reason directly within the latent visual space.
  arXiv  Detail & Related papers  (2025-11-26T13:46:39Z)
• Explaining multimodal LLMs via intra-modal token interactions [55.27436637894534]
  Multimodal Large Language Models (MLLMs) have achieved remarkable success across diverse vision-language tasks, yet their internal decision-making mechanisms remain insufficiently understood.<n>We propose enhancing interpretability by leveraging intra-modal interaction.
  arXiv  Detail & Related papers  (2025-09-26T14:39:13Z)
• BASIC: Boosting Visual Alignment with Intrinsic Refined Embeddings in Multimodal Large Language Models [10.16893890191528]
  Multimodal Large Language Models (MLLMs) achieve visual understanding by using a vision projector to bridge well-pretrained vision encoders and large language models.<n>Current alignment approaches treat visual embeddings as contextual cues and merely apply auto-regressive supervision to textual outputs.<n>We propose BASIC, a method that utilizes refined visual embeddings within the LLM as supervision to directly guide the projector in generating initial visual embeddings.
  arXiv  Detail & Related papers  (2025-08-09T09:00:45Z)
• ATAS: Any-to-Any Self-Distillation for Enhanced Open-Vocabulary Dense Prediction [7.353998772647553]
  Any-to-Any Self-Distillation (ATAS) is a novel approach that simultaneously enhances semantic coherence and fine-grained alignment.<n>ATAS achieves substantial performance gains on open-vocabulary object detection and semantic segmentation benchmarks.
  arXiv  Detail & Related papers  (2025-06-10T10:40:10Z)
• Cantor: Inspiring Multimodal Chain-of-Thought of MLLM [83.6663322930814]
  We argue that converging visual context acquisition and logical reasoning is pivotal for tackling visual reasoning tasks. We propose an innovative multimodal CoT framework, termed Cantor, characterized by a perception-decision architecture. Our experiments demonstrate the efficacy of the proposed framework, showing significant improvements in multimodal CoT performance.
  arXiv  Detail & Related papers  (2024-04-24T17:59:48Z)

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.