PointThinker：面向多模态大模型的点激励并行思维
PointThinker: Point-Incentivized Parallel Thinking for Multimodal Large Language Model
把「先列出多个视觉关键点、再以每个点为起点展开独立推理链」作为并行思维链的组织方式，并配合点级（dense）奖励的强化学习（GPPO）做自验证，可直接迁移到多模态推理、幻觉抑制等需要多路径推理的视觉语言任务中。
原文依据 We guide the model to list multiple visual key points and develop an independent reasoning path for each.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论本文探索多模态大模型（MLLM）中的并行思维，旨在通过多条多样化的推理路径改进思维链（CoT）。我们引导模型列出多个视觉关键点，并为每个关键点发展一条独立的推理路径。因此，我们将该方法称为 PointThinker，其特点是每条思维路径都以一个点作为起点。PointThinker 具有两个关键优势。（1）它放大了并行思维的收益。并行思维天然受益于多条推理路径，而显式列出关键点通过消除冗余、促进路径多样性进一步放大了这些收益，使模型能够从更多样的视角探索问题。（2）它使用一种新颖的稠密（逐点）奖励进行强化学习。我们观察到，在并行思维过程中，有些点是有帮助的，而另一些是无效的，但主流方法却给它们分配相同的奖励。因此，我们提出对同一条思维链内的不同点分配差异化的奖励。这通过一种名为分组点策略优化（Group Points Policy Optimization，GPPO）的自验证机制实现，该机制结合了 rollout 级与点级验证来进行奖励分配。在 HallusionBench 等具有挑战性的基准上，PointThinker 达到 58.7% 的准确率，提升了推理质量与答案准确率。实验结果表明，以点为起点的并行思维能够提升性能，而 GPPO 进一步带来了非平凡的增益。
英文原文摘要 1522 字符
This paper explores parallel thinking for Multi-modal Large Language Models (MLLMs), aiming to improve Chain-of-Thought (CoT) through multiple diverse reasoning paths. We guide the model to list multiple visual key points and develop an independent reasoning path for each. Therefore, we term this method PointThinker, which is characterized by starting each thinking path with a point. PointThinker offers two key advantages. (1) It amplifies the benefits of parallel thinking. While parallel thinking naturally benefits from multiple reasoning paths, explicitly listing key points further amplifies these benefits by eliminating redundancy and promoting path diversity, enabling the model to explore problems from more varied perspectives. (2) It uses a novel dense (point-wise) reward for reinforcement learning. We observe that during parallel thinking, some points are helpful while others are invalid, yet popular methods assign them the same rewards. Therefore, we propose allocating differentiated rewards to different points within the same chain-of-thought. This is implemented via a self-verification mechanism called Group Points Policy Optimization (GPPO), which combines rollout-level and point-level validation for reward assignment. On challenging benchmarks such as HallusionBench, PointThinker achieves 58.7% accuracy, improving reasoning quality and answer accuracy. Experimental results demonstrate that parallel thinking with point improves performance, and GPPO further contributes non-trivial gains.