analiza aiqbench
The paper proposes a large synthetic dataset of visual logic puzzles, generated by a rule-to-image pipeline, for training vision-language models with reinforcement learning (GRPO and binary correctness rewards). The reported gains are on logical reasoning, with some transfer to algebra, arithmetic, and geometry, but the abstract gives no quantitative results, so the size of the improvement is unclear. For readers, the useful part is the approach: programmatically generated, verifiable training data as an alternative to scarce human-labelled multimodal reasoning data.
VisualSphinx: Large-Scale Synthetic Vision Logic Puzzles for RL
arXiv:2505.23977v2 Announce Type: replace-cross Abstract: Vision language models (VLMs) are expected to perform effective multimodal reasoning and make logically coherent decisions, which is critical to tasks such as diagram understanding and spatial problem solving. However, current VLM reasoning lacks large-scale and well-structured training datasets. To bridge this gap, we propose VisualSphinx, a first-of-its-kind large-scale synthetic visual logical reasoning training dataset. To tackle the challenge of image synthesis with grounded answers, we propose a rule-to-image synthesis pipeline that extracts and expands puzzle rules from seed questions and generates executable rendering programs for puzzle assembly. Using standard GRPO with binary correctness rewards as a validation recipe, experiments show that VLMs trained on VisualSphinx benefit from the logical coherence and readability of our dataset and exhibit improved performance on logical reasoning tasks. The enhanced reasoning capabilities developed from VisualSphinx also benefit other reasoning tasks such as algebraic reasoning, arithmetic reasoning and geometry reasoning.
Abstract:Vision language models (VLMs) are expected to perform effective multimodal reasoning and make logically coherent decisions, which is critical to tasks such as diagram understanding and spatial problem solving. However, current VLM reasoning lacks large-scale and well-structured training datasets. To bridge this gap, we propose VisualSphinx, a first-of-its-kind large-scale synthetic visual logical reasoning training dataset. To tackle the challenge of image synthesis with grounded answers, we propose a rule-to-image synthesis pipeline that extracts and expands puzzle rules from seed questions and generates executable rendering programs for puzzle assembly. Using standard GRPO with binary correctness rewards as a validation recipe, experiments show that VLMs trained on VisualSphinx benefit from the logical coherence and readability of our dataset and exhibit improved performance on logical reasoning tasks. The enhanced reasoning capabilities developed from VisualSphinx also benefit other reasoning tasks such as algebraic reasoning, arithmetic reasoning and geometry reasoning.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2505.23977 [cs.CV]
(or arXiv:2505.23977v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2505.23977 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 246 cuvinte