Enregistré dans:
| Auteurs principaux: | , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | https://arxiv.org/abs/2601.08010 |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866917198610563072 |
|---|---|
| author | Li, Chaoyu Barua, Deeparghya Dutta Tao, Fei Fazli, Pooyan |
| author_facet | Li, Chaoyu Barua, Deeparghya Dutta Tao, Fei Fazli, Pooyan |
| contents | Vision-language models achieve strong performance across a wide range of multimodal understanding and reasoning tasks, yet their multi-step reasoning remains unstable. Repeated sampling over the same input often produces divergent reasoning trajectories and inconsistent final predictions. To address this, we introduce two complementary approaches inspired by test-time scaling: (1) CASHEW, an inference-time framework that stabilizes reasoning by iteratively aggregating multiple candidate trajectories into higher-quality reasoning traces, with explicit visual verification filtering hallucinated steps and grounding reasoning in visual evidence, and (2) CASHEW-RL, a learned variant that internalizes this aggregation behavior within a single model. CASHEW-RL is trained using Group Sequence Policy Optimization (GSPO) with a composite reward that encourages correct answers grounded in minimal yet sufficient visual evidence, while adaptively allocating reasoning effort based on task difficulty. This training objective enables robust self-aggregation at inference. Extensive experiments on 13 image understanding, video understanding, and video reasoning benchmarks show significant performance improvements, including gains of up to +23.6 percentage points on ScienceQA and +8.1 percentage points on EgoSchema. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_08010 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | CASHEW: Stabilizing Multimodal Reasoning via Iterative Trajectory Aggregation Li, Chaoyu Barua, Deeparghya Dutta Tao, Fei Fazli, Pooyan Computer Vision and Pattern Recognition Vision-language models achieve strong performance across a wide range of multimodal understanding and reasoning tasks, yet their multi-step reasoning remains unstable. Repeated sampling over the same input often produces divergent reasoning trajectories and inconsistent final predictions. To address this, we introduce two complementary approaches inspired by test-time scaling: (1) CASHEW, an inference-time framework that stabilizes reasoning by iteratively aggregating multiple candidate trajectories into higher-quality reasoning traces, with explicit visual verification filtering hallucinated steps and grounding reasoning in visual evidence, and (2) CASHEW-RL, a learned variant that internalizes this aggregation behavior within a single model. CASHEW-RL is trained using Group Sequence Policy Optimization (GSPO) with a composite reward that encourages correct answers grounded in minimal yet sufficient visual evidence, while adaptively allocating reasoning effort based on task difficulty. This training objective enables robust self-aggregation at inference. Extensive experiments on 13 image understanding, video understanding, and video reasoning benchmarks show significant performance improvements, including gains of up to +23.6 percentage points on ScienceQA and +8.1 percentage points on EgoSchema. |
| title | CASHEW: Stabilizing Multimodal Reasoning via Iterative Trajectory Aggregation |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.08010 |