Metraipolis › archive › dc_2e453cc2d71d
TTRSD: Test-Time Reinforcement Learning with Self-Distillation for Vision-Language Models — arXiv:2609.33414
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.33414
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T21:01:25.473Z
- SHA-256 of the archived text
7016db533dba01906a357b5e3ba6338223a1cf503178d9afc7426417f86e46fd- Stable URL of this page
https://metraipolis.com/documento/dc_2e453cc2d71d
How to cite
TTRSD: Test-Time Reinforcement Learning with Self-Distillation for Vision-Language Models — arXiv:2609.33414 — arXiv (en), https://arxiv.org/abs/2609.33414 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2e453cc2d71d (Metraipolis archive dc_2e453cc2d71d, sha256 7016db533dba01906a357b5e3ba6338223a1cf503178d9afc7426417f86e46fd)
BibTeX
@misc{dc2e453cc2d71d,
title = {TTRSD: Test-Time Reinforcement Learning with Self-Distillation for Vision-Language Models — arXiv:2609.33414},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.33414},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2e453cc2d71d; sha256 7016db533dba01906a357b5e3ba6338223a1cf503178d9afc7426417f86e46fd},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.33414. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Test-time reinforcement learning enables vision-language models (VLMs) to adapt using unlabeled inputs. However, repeated sampling under fixed visual conditions can reinforce shared perceptual errors, while sequence-level rewards fail to isolate visual perception the foundational bottleneck that anchors multimodal reasoning risking the degradation of pre-trained reasoning capabilities. We propose TTRSD, a test-time reinforcement learning framework combining multi-view answer-level self-distillation with visual contrastive token selection. A shared policy aggregates teacher predictions across original, cropped, and downsampled views into an answer distribution. Student trajectories generated from the original image receive rewards based on the support for their final answers in this distribution. To allocate this feedback precisely toward perceptual bottlenecks, we compare the log-probabilities of the same sampled tokens under original and visually ablated inputs while holding their textual prefixes fixed, selecting visually sensitive positions for policy-gradient updates. TTRSD separates update direction, determined by group-relative advantages, from update position, determined by visual sensitivity, without requiring ground-truth labels, external verifiers, or a separate teacher. With only 20 unlabeled adaptation samples, TTRSD improves performance across seven benchmarks and three VLMs, raising InternVL3-2B's MMMU accuracy from 35.79% to 49.32%(+13.53%), demonstrating cross-dataset generalization while preserving inherent reasoning integrity. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.33414 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T20:29:02.650Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).