Multimodal Latent Reasoning via Hierarchical Visual Cues Injection
This addresses the reasoning limitations of multimodal AI systems, offering a more robust approach for applications requiring complex scene understanding, though it appears incremental as an extension of existing transformer-based methods.
The paper tackles the problem of inefficient and hallucination-prone reasoning in multimodal large language models by proposing HIVE, a framework that enables grounded, multi-step inference in latent space through hierarchical visual cues injection, achieving effective test-time scaling and enhanced understanding of complex scenes.
The advancement of multimodal large language models (MLLMs) has enabled impressive perception capabilities. However, their reasoning process often remains a "fast thinking" paradigm, reliant on end-to-end generation or explicit, language-centric chains of thought (CoT), which can be inefficient, verbose, and prone to hallucination. This work posits that robust reasoning should evolve within a latent space, integrating multimodal signals seamlessly. We propose multimodal latent reasoning via HIerarchical Visual cuEs injection (\emph{HIVE}), a novel framework that instills deliberate, "slow thinking" without depending on superficial textual rationales. Our method recursively extends transformer blocks, creating an internal loop for iterative reasoning refinement. Crucially, it injectively grounds this process with hierarchical visual cues from global scene context to fine-grained regional details directly into the model's latent representations. This enables the model to perform grounded, multi-step inference entirely in the aligned latent space. Extensive evaluations demonstrate that test-time scaling is effective when incorporating vision knowledge, and that integrating hierarchical information significantly enhances the model's understanding of complex scenes.