When Models Judge Themselves: Unsupervised Self-Evolution for Multimodal Reasoning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wu, Zhengxian, Shi, Kai, Zhang, Chuanrui, Liao, Zirui, Yang, Jun, Yang, Ni, Peng, Qiuying, Zhang, Luyuan, Xu, Hangrui, Su, Tianhuang, Yang, Zhenyu, Lu, Haonan, Wang, Haoqian
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917359081488384
author Wu, Zhengxian
Shi, Kai
Zhang, Chuanrui
Liao, Zirui
Yang, Jun
Yang, Ni
Peng, Qiuying
Zhang, Luyuan
Xu, Hangrui
Su, Tianhuang
Yang, Zhenyu
Lu, Haonan
Wang, Haoqian
author_facet Wu, Zhengxian
Shi, Kai
Zhang, Chuanrui
Liao, Zirui
Yang, Jun
Yang, Ni
Peng, Qiuying
Zhang, Luyuan
Xu, Hangrui
Su, Tianhuang
Yang, Zhenyu
Lu, Haonan
Wang, Haoqian
contents Recent progress in multimodal large language models has led to strong performance on reasoning tasks, but these improvements largely rely on high-quality annotated data or teacher-model distillation, both of which are costly and difficult to scale. To address this, we propose an unsupervised self-evolution training framework for multimodal reasoning that achieves stable performance improvements without using human-annotated answers or external reward models. For each input, we sample multiple reasoning trajectories and jointly model their within group structure. We use the Actor's self-consistency signal as a training prior, and introduce a bounded Judge based modulation to continuously reweight trajectories of different quality. We further model the modulated scores as a group level distribution and convert absolute scores into relative advantages within each group, enabling more robust policy updates. Trained with Group Relative Policy Optimization (GRPO) on unlabeled data, our method consistently improves reasoning performance and generalization on five mathematical reasoning benchmarks, offering a scalable path toward self-evolving multimodal models. The code are available at https://github.com/OPPO-Mente-Lab/LLM-Self-Judge.
format Preprint
id arxiv_https___arxiv_org_abs_2603_21289
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle When Models Judge Themselves: Unsupervised Self-Evolution for Multimodal Reasoning
Wu, Zhengxian
Shi, Kai
Zhang, Chuanrui
Liao, Zirui
Yang, Jun
Yang, Ni
Peng, Qiuying
Zhang, Luyuan
Xu, Hangrui
Su, Tianhuang
Yang, Zhenyu
Lu, Haonan
Wang, Haoqian
Computer Vision and Pattern Recognition
Artificial Intelligence
Recent progress in multimodal large language models has led to strong performance on reasoning tasks, but these improvements largely rely on high-quality annotated data or teacher-model distillation, both of which are costly and difficult to scale. To address this, we propose an unsupervised self-evolution training framework for multimodal reasoning that achieves stable performance improvements without using human-annotated answers or external reward models. For each input, we sample multiple reasoning trajectories and jointly model their within group structure. We use the Actor's self-consistency signal as a training prior, and introduce a bounded Judge based modulation to continuously reweight trajectories of different quality. We further model the modulated scores as a group level distribution and convert absolute scores into relative advantages within each group, enabling more robust policy updates. Trained with Group Relative Policy Optimization (GRPO) on unlabeled data, our method consistently improves reasoning performance and generalization on five mathematical reasoning benchmarks, offering a scalable path toward self-evolving multimodal models. The code are available at https://github.com/OPPO-Mente-Lab/LLM-Self-Judge.
title When Models Judge Themselves: Unsupervised Self-Evolution for Multimodal Reasoning
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2603.21289