Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2602.00574 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917238513074176 |
|---|---|
| author | Shao, Yifei Zhou, Kun Xu, Ziming Quamar, Mohammad Atif Hao, Shibo Wang, Zhen Hu, Zhiting Huang, Biwei |
| author_facet | Shao, Yifei Zhou, Kun Xu, Ziming Quamar, Mohammad Atif Hao, Shibo Wang, Zhen Hu, Zhiting Huang, Biwei |
| contents | We study how to extend chain-of-thought (CoT) beyond language to better handle multimodal reasoning. While CoT helps LLMs and VLMs articulate intermediate steps, its text-only form often fails on vision-intensive problems where key intermediate states are inherently visual. We introduce modal-mixed CoT, which interleaves textual tokens with compact visual sketches represented as latent embeddings. To bridge the modality gap without eroding the original knowledge and capability of the VLM, we use the VLM itself as an encoder and train the language backbone to reconstruct its own intermediate vision embeddings, to guarantee the semantic alignment of the visual latent space. We further attach a diffusion-based latent decoder, invoked by a special control token and conditioned on hidden states from the VLM. In this way, the diffusion head carries fine-grained perceptual details while the VLM specifies high-level intent, which cleanly disentangles roles and reduces the optimization pressure of the VLM. Training proceeds in two stages: supervised fine-tuning on traces that interleave text and latents with a joint next-token and latent-reconstruction objective, followed by reinforcement learning that teaches when to switch modalities and how to compose long reasoning chains. Extensive experiments across 11 diverse multimodal reasoning tasks, demonstrate that our method yields better performance than language-only and other CoT methods. Our code will be publicly released. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_00574 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Learning Modal-Mixed Chain-of-Thought Reasoning with Latent Embeddings Shao, Yifei Zhou, Kun Xu, Ziming Quamar, Mohammad Atif Hao, Shibo Wang, Zhen Hu, Zhiting Huang, Biwei Artificial Intelligence We study how to extend chain-of-thought (CoT) beyond language to better handle multimodal reasoning. While CoT helps LLMs and VLMs articulate intermediate steps, its text-only form often fails on vision-intensive problems where key intermediate states are inherently visual. We introduce modal-mixed CoT, which interleaves textual tokens with compact visual sketches represented as latent embeddings. To bridge the modality gap without eroding the original knowledge and capability of the VLM, we use the VLM itself as an encoder and train the language backbone to reconstruct its own intermediate vision embeddings, to guarantee the semantic alignment of the visual latent space. We further attach a diffusion-based latent decoder, invoked by a special control token and conditioned on hidden states from the VLM. In this way, the diffusion head carries fine-grained perceptual details while the VLM specifies high-level intent, which cleanly disentangles roles and reduces the optimization pressure of the VLM. Training proceeds in two stages: supervised fine-tuning on traces that interleave text and latents with a joint next-token and latent-reconstruction objective, followed by reinforcement learning that teaches when to switch modalities and how to compose long reasoning chains. Extensive experiments across 11 diverse multimodal reasoning tasks, demonstrate that our method yields better performance than language-only and other CoT methods. Our code will be publicly released. |
| title | Learning Modal-Mixed Chain-of-Thought Reasoning with Latent Embeddings |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2602.00574 |