Saved in:
Bibliographic Details
Main Authors: Qiu, Xinyu, Jia, Heng, Zeng, Zhengwen, Shen, Shuheng, Meng, Changhua, Yang, Yi, Zhu, Linchao
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.01483
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909981101522944
author Qiu, Xinyu
Jia, Heng
Zeng, Zhengwen
Shen, Shuheng
Meng, Changhua
Yang, Yi
Zhu, Linchao
author_facet Qiu, Xinyu
Jia, Heng
Zeng, Zhengwen
Shen, Shuheng
Meng, Changhua
Yang, Yi
Zhu, Linchao
contents Parallel test-time scaling typically trains separate generation and verification models, incurring high training and inference costs. We propose Advantage Decoupled Preference Optimization (ADPO), a unified reinforcement learning framework that jointly learns answer generation and self-verification within a single policy. ADPO introduces two innovations: a preference verification reward improving verification capability and a decoupled optimization mechanism enabling synergistic optimization of generation and verification. Specifically, the preference verification reward computes mean verification scores from positive and negative samples as decision thresholds, providing positive feedback when prediction correctness aligns with answer correctness. Meanwhile, the advantage decoupled optimization computes separate advantages for generation and verification, applies token masks to isolate gradients, and combines masked GRPO objectives, preserving generation quality while calibrating verification scores. ADPO achieves up to +34.1% higher verification AUC and -53.5% lower inference time, with significant gains of +2.8%/+1.4% accuracy on MathVista/MMMU, +1.9 cIoU on ReasonSeg, and +1.7%/+1.0% step success rate on AndroidControl/GUI Odyssey.
format Preprint
id arxiv_https___arxiv_org_abs_2601_01483
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Unified Generation and Self-Verification for Vision-Language Models via Advantage Decoupled Preference Optimization
Qiu, Xinyu
Jia, Heng
Zeng, Zhengwen
Shen, Shuheng
Meng, Changhua
Yang, Yi
Zhu, Linchao
Computer Vision and Pattern Recognition
Parallel test-time scaling typically trains separate generation and verification models, incurring high training and inference costs. We propose Advantage Decoupled Preference Optimization (ADPO), a unified reinforcement learning framework that jointly learns answer generation and self-verification within a single policy. ADPO introduces two innovations: a preference verification reward improving verification capability and a decoupled optimization mechanism enabling synergistic optimization of generation and verification. Specifically, the preference verification reward computes mean verification scores from positive and negative samples as decision thresholds, providing positive feedback when prediction correctness aligns with answer correctness. Meanwhile, the advantage decoupled optimization computes separate advantages for generation and verification, applies token masks to isolate gradients, and combines masked GRPO objectives, preserving generation quality while calibrating verification scores. ADPO achieves up to +34.1% higher verification AUC and -53.5% lower inference time, with significant gains of +2.8%/+1.4% accuracy on MathVista/MMMU, +1.9 cIoU on ReasonSeg, and +1.7%/+1.0% step success rate on AndroidControl/GUI Odyssey.
title Unified Generation and Self-Verification for Vision-Language Models via Advantage Decoupled Preference Optimization
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2601.01483