arXiv cs.CL· Seonghoon Yu, Dongjun Nam, Byung-Kwan Lee, Jeany Son·· 11 小时前AI 评分31
Hide to See：面向 VLM 蒸馏中视觉锚定思考的推理前缀掩码方法
Hide to See: Reasoning-prefix Masking for Visual-anchored Thinking in VLM Distillation
AI 导读
研究者提出一种 think-answer 蒸馏框架，通过掩码学生模型的显著推理前缀，迫使其在推理过程中更多依赖视觉证据，以缓解长推理链中的视觉遗忘问题。方法包含 token 级显著推理前缀掩码和按师生分布差异衡量的自适应掩码预算调度，并在蒸馏阶段替代标准因果掩码。在多模态推理基准上，该方法优于近期开源 VLM、VLM 蒸馏与自蒸馏方法，分析显示学生思考过程中的视觉利用率得到提升。
正文 · 原文
Abstract:Recent think-answer approaches in VLMs, such as Qwen3-VL-Thinking, boost reasoning performance by leveraging intermediate thinking steps before the final answer, but their computational cost becomes substantial, especially for larger VLMs. To distill such capabilities into compact think-answer VLMs, a primary objective is to improve the student's ability to utilize visual evidence throughout its reasoning trace, as long think-answer traces suffer from visual forgetting issues. To this end, we introduce a novel think-answer distillation framework that encourages the student to anchor its thinking on visual information by masking the student's salient reasoning prefixes. To compensate for such masked textual cues, the student is encouraged to rely more on visual evidence as an alternative source of information during distillation. Our masking strategies include: 1) token-wise salient reasoning-prefix masking, which masks high-influence reasoning prefixes selectively for each next-token prediction, and 2) self-paced masking budget scheduling, which gradually increases the masking scale according to distillation difficulty, measured by the discrepancy between teacher--student distributions. In the distillation phase, the student is guided by our salient reasoning-prefix mask, which blocks both future tokens and salient reasoning cues, in place of the standard causal mask used for auto-regressive language modeling. Experimental results show that our approach outperforms recent open-source VLMs, VLM distillation, and self-distillation methods on multimodal reasoning benchmarks, while further analyzes confirm enhanced visual utilization along the student thinking process.
| Comments: | Accepted to NeurIPS 2026 |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.11651 [cs.CV] |
| (or arXiv:2605.11651v5 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2605.11651 arXiv-issued DOI via DataCite |
Submission history
From: Seonghoon Yu [view email]
[v1]
Tue, 12 May 2026 07:14:04 UTC (6,058 KB)
[v2]
Wed, 13 May 2026 01:49:55 UTC (6,058 KB)
[v3]
Fri, 15 May 2026 06:49:33 UTC (6,058 KB)
[v4]
Tue, 26 May 2026 04:36:02 UTC (6,058 KB)
[v5]
Fri, 6 Nov 2026 08:21:59 UTC (5,866 KB)
来源：arXiv cs.CL · arxiv.org