| title | Beyond Scalar Rewards by Internalizing Reasoning into Score Distributions |
|---|---|
| authors | Xin Jin, Huanqia Cai, Zhen Li, Zechao Zhan, Dengyang Jiang, Aiming Hao, Yuming Jiang, Chunle Guo, Peng Gao, Ming-Ming Cheng, Steven C. H. Hoi |
| date | 2026-06-08 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2606.09076 |
| affiliation | Tongyi-MAI |
| upvotes | 61 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf |
| abstract | Reward models are central to text-to-image post-training, but visual preference is subjective and better represented as a distribution over rubric scores than as a deterministic scalar. Existing scalar, score-token, and pairwise reward models over-compress uncertainty and fine-grained score differences, while reasoning-based generative rewards provide stronger judgments but are costly to deploy and difficult to use as direct optimization signals. We propose Z-Reward, a teacher-student reward modeling framework that decouples reasoning-heavy judgment from efficient reward deployment. The teacher is a large VLM that uses reasoning to infer rubric-aligned score distributions, and is trained with Group-wise Direct Score Optimization (GDSO), which combines policy-gradient rewards from distribution expectations with direct pointwise and pairwise supervision on score distributions and score gaps. The student is trained with Reasoning-Internalized Score Distillation (RISD), which transfers the teacher's reasoning-conditioned score distribution into a compact VLM without requiring explicit reasoning chains at inference time. On our internally annotated evaluation set, the 27B GDSO teacher reaches 89.6% human preference accuracy, outperforming SFT, RewardDance, and GRPO, while the 9B RISD student reaches 88.6%, outperforming the OPD baseline and closely matching the larger teacher. We further show that Z-Reward can serve as a differentiable reward signal for text-to-image optimization, yielding a 41.3% net human-preference improvement over the SFT baseline. |
cite
@misc{jin2026beyond,
title={Beyond Scalar Rewards by Internalizing Reasoning into Score Distributions},
author={Xin Jin and Huanqia Cai and Zhen Li and Zechao Zhan and Dengyang Jiang and Aiming Hao and Yuming Jiang and Chunle Guo and Peng Gao and Ming-Ming Cheng and Steven C. H. Hoi},
year={2026},
eprint={2606.09076},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2606.09076}
}