arXiv cs.AI· Wei Cai, Jian Zhao, Yuchen Yuan, Xuelong Li·· 12 小时前AI 评分31
CoRAL：用扩散-自回归混合框架突破多模态推理的顺序性瓶颈
Beyond the Sequentiality Bottleneck: A Hybrid Diffusion-Autoregressive Framework for Efficient Multimodal Reasoning
AI 导读
研究者提出混合框架 CoRAL，通过非对称 Planner-Executor（APE）架构让扩散语言模型 Planner 并行生成 K 条推理轨迹，再由大型自回归模型 Executor 选出最优方案并合成最终答案。
正文 · 原文
Abstract:Autoregressive models (ARMs) face a "Sequentiality Bottleneck" in reasoning, while diffusion language models (DLMs) offer parallelism but lack semantic coherence. To addresses both limitations, we propose Collaborative Reasoning Asymmetric Learning (CoRAL), a hybrid framework via an Asymmetric Planner-Executor (APE) architecture: a DLM Planner generates K diverse reasoning trajectories in parallel, while a large ARM Executor selects the optimal plan and synthesizes the final answer. To train the Planner effectively, we introduce Reward-Guided Denoising for CoT (RDC), which integrates outcome-based rewards into the diffusion process to steer it toward goal-directed reasoning. Experiments on vision-language benchmarks show that CoRAL significantly outperforms prior diffusion-based methods and achieves competitive performance against strong ARMs, achieving both superior efficiency and quality.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.28584 [cs.AI] |
| (or arXiv:2610.28584v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.28584 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Wei Cai [view email]
[v1]
Tue, 20 Oct 2026 17:39:41 UTC (2,429 KB)
来源：arXiv cs.AI · arxiv.org