Metraipolis › archive › dc_b128366ac839

Answer Correction with Verifiable Rewards: A Gemma-Family Study of Multi-View Driving QA — arXiv:2611.00828

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.00828
License (as archived)
CC0-1.0
Archived
2026-11-03T11:24:57.359Z
SHA-256 of the archived text
62ca9285b353bd105651ab9bf8aaff40524c20b44745c000ca359af76c7c68b4
Stable URL of this page
https://metraipolis.com/documento/dc_b128366ac839

How to cite

Answer Correction with Verifiable Rewards: A Gemma-Family Study of Multi-View Driving QA — arXiv:2611.00828 — arXiv (en), https://arxiv.org/abs/2611.00828 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_b128366ac839 (Metraipolis archive dc_b128366ac839, sha256 62ca9285b353bd105651ab9bf8aaff40524c20b44745c000ca359af76c7c68b4)
BibTeX
@misc{dcb128366ac839,
  title = {Answer Correction with Verifiable Rewards: A Gemma-Family Study of Multi-View Driving QA — arXiv:2611.00828},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.00828},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_b128366ac839; sha256 62ca9285b353bd105651ab9bf8aaff40524c20b44745c000ca359af76c7c68b4},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.00828. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Multi-camera driving question answering provides a discrete test of vision-language models, but post-training gains can reflect changed selected answers or improved answer extraction. We investigate this distinction within the Gemma model family using archived, task-linked WaymoQA evaluations. A zero-shot diagnostic on 5,519 test questions spans effective-small, mixture-of-experts, unified, and dense checkpoints, with serving differences disclosed rather than interpreted as architecture effects. The strongest controlled evidence is a matched-harness Gemma-4-12B base/GRPO pair: accuracy rises from 73.78% to 75.43%, a gain of 1.65 percentage points with a scene-cluster bootstrap interval of [+0.84, +2.41]. There are 215 improvements and 124 regressions; 213 improvements start from an extractable but incorrect base answer, and only two recover previously unextractable responses. Crossing correctness with extraction state shows why correct noncompliant outputs cannot be counted as prediction errors. Category counts reveal a critical safety asymmetry: routine perception gains (+5.93 pp) mask a regression in safety-critical hazard prediction (-1.82 pp), demonstrating that unweighted outcome rewards can trade off rare edge cases for aggregate accuracy. Reward saturation describes sampled training dynamics rather than an architectural ceiling. A separate archived Gemma-31B thinking comparison yields no detected average improvement, with uncertainty that does not establish equivalence. These results support a modest answer-selection gain for the recorded tuning configuration and a reproducible diagnostic accounting, without attributing corrected labels to specific grounding mechanisms or claiming superiority over matched supervised tuning.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.00828 — 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 2026-11-03T11:02:37.792Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).