SynthRL: Scaling Visual Reasoning with Verifiable Data Synthesis

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wu, Zijian, Ni, Jinjie, Liu, Xiangyan, Liu, Zichen, Yan, Hang, Shieh, Michael Qizhe
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916774455279616
author Wu, Zijian
Ni, Jinjie
Liu, Xiangyan
Liu, Zichen
Yan, Hang
Shieh, Michael Qizhe
author_facet Wu, Zijian
Ni, Jinjie
Liu, Xiangyan
Liu, Zichen
Yan, Hang
Shieh, Michael Qizhe
contents Vision-language models (VLMs) trained via reinforcement learning with verifiable reward (RLVR) have shown notable progress in scaling test-time compute effectively. In this work, we investigate how synthesized RL data can further improve RLVR. To this end, we propose \textbf{SynthRL}-a scalable and guaranteed pipeline for automatic data scaling in reasoning-oriented RL training. SynthRL comprises three key stages: (1) selecting seed questions with appropriate distribution, (2) augmenting them into more challenging variants while preserving the original answers, and (3) a guaranteed verification stage that ensures near-perfect correctness and difficulty enhancement. Our empirical experiments demonstrate SynthRL's scalability and effectiveness. When applied to the MMK12 dataset, SynthRL synthesizes over 3.3K additional verifiable, challenging questions from approximately 8K seed samples. Models trained with our synthesized data achieve consistent gains across five out-of-domain visual math reasoning benchmarks, with a significant improvement over baseline models trained on seed data alone. Notably, detailed analysis reveals that the gains are more pronounced on the most challenging evaluation samples, highlighting SynthRL's effectiveness in eliciting deeper and more complex reasoning patterns.
format Preprint
id arxiv_https___arxiv_org_abs_2506_02096
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SynthRL: Scaling Visual Reasoning with Verifiable Data Synthesis
Wu, Zijian
Ni, Jinjie
Liu, Xiangyan
Liu, Zichen
Yan, Hang
Shieh, Michael Qizhe
Machine Learning
Computation and Language
Computer Vision and Pattern Recognition
Vision-language models (VLMs) trained via reinforcement learning with verifiable reward (RLVR) have shown notable progress in scaling test-time compute effectively. In this work, we investigate how synthesized RL data can further improve RLVR. To this end, we propose \textbf{SynthRL}-a scalable and guaranteed pipeline for automatic data scaling in reasoning-oriented RL training. SynthRL comprises three key stages: (1) selecting seed questions with appropriate distribution, (2) augmenting them into more challenging variants while preserving the original answers, and (3) a guaranteed verification stage that ensures near-perfect correctness and difficulty enhancement. Our empirical experiments demonstrate SynthRL's scalability and effectiveness. When applied to the MMK12 dataset, SynthRL synthesizes over 3.3K additional verifiable, challenging questions from approximately 8K seed samples. Models trained with our synthesized data achieve consistent gains across five out-of-domain visual math reasoning benchmarks, with a significant improvement over baseline models trained on seed data alone. Notably, detailed analysis reveals that the gains are more pronounced on the most challenging evaluation samples, highlighting SynthRL's effectiveness in eliciting deeper and more complex reasoning patterns.
title SynthRL: Scaling Visual Reasoning with Verifiable Data Synthesis
topic Machine Learning
Computation and Language
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.02096