CrystaL: Spontaneous Emergence of Visual Latents in MLLMs
This work addresses a bottleneck in MLLMs for researchers and practitioners by improving visual reasoning without auxiliary annotations, though it is incremental as it builds on existing latent CoT methods.
The paper tackled the problem of limited guidance for preserving critical visual information in intermediate latent states of multimodal large language models (MLLMs) using latent Chain-of-Thought methods, and proposed CrystaL, a single-stage framework that aligns attention patterns and prediction distributions across intact and corrupted image paths, resulting in substantial gains in fine-grained visual understanding on perception-intensive benchmarks.
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.