analiza aiqbench
This research introduces EASE, a method that enhances Reinforcement Learning with Verifiable Rewards (RLVR) by supervising visual attention on specific evidence regions rather than relying solely on final answer rewards. By guiding the model's focus during training on high-reward trajectories, EASE significantly reduces hallucinations and improves performance in visual grounding tasks across multiple Qwen VLM benchmarks. The approach demonstrates that process supervision with privileged labels can effectively align multimodal models with actual visual evidence without compromising inference efficiency.
Attend to Evidence: Evidence-Anchored Spatial Attention Supervision for Multimodal RLVR
arXiv:2605.30912v2 Announce Type: replace-cross Abstract: Reinforcement learning with verifiable rewards (RLVR) improves vision-language models (VLMs) by optimizing outcome rewards derived from final answers. However, such outcome-only rewards do not tell the model which image regions justify an answer. For questions that require visual grounding, these rewards cannot distinguish responses supported by relevant visual evidence from those produced by language-prior shortcuts or lucky guesses. We introduce EASE (Evidence-Anchored Spatial Attention), which augments multimodal RLVR with visual-evidence process supervision. EASE converts annotated evidence regions into a smoothed visual-token target and uses it to guide response-to-image attention during RL training, but only on high-reward trajectories. The annotations are used solely as privileged training labels, while inference requires only the original image and question. Across Qwen2.5-VL-7B, Qwen3-VL-4B, and Qwen3-VL-8B, EASE raises average scores over DAPO by 2.5 to 3.1 points on perception, hallucination, visual math, and multimodal reasoning benchmarks. Diagnostics and ablations show that EASE better aligns visual attention with annotated evidence regions.
Abstract:Reinforcement learning with verifiable rewards (RLVR) improves vision-language models (VLMs) by optimizing outcome rewards derived from final answers. However, such outcome-only rewards do not tell the model which image regions justify an answer. For questions that require visual grounding, these rewards cannot distinguish responses supported by relevant visual evidence from those produced by language-prior shortcuts or lucky guesses. We introduce EASE (Evidence-Anchored Spatial Attention), which augments multimodal RLVR with visual-evidence process supervision. EASE converts annotated evidence regions into a smoothed visual-token target and uses it to guide response-to-image attention during RL training, but only on high-reward trajectories. The annotations are used solely as privileged training labels, while inference requires only the original image and question. Across Qwen2.5-VL-7B, Qwen3-VL-4B, and Qwen3-VL-8B, EASE raises average scores over DAPO by 2.5 to 3.1 points on perception, hallucination, visual math, and multimodal reasoning benchmarks. Diagnostics and ablations show that EASE better aligns visual attention with annotated evidence regions.
Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
arXiv:2605.30912 [cs.CV]
(or arXiv:2605.30912v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2605.30912 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 · 253 cuvinte
Subiecte conexe
Spatially Grounded Concept Bottleneck Models via Part-Factorized Attention
arXiv · 29 iul. 2026
Mind What Matters for Reasoning: Aligning Cross-Modal Attention via Selective Probability Mass Concentration
arXiv · 25 sept. 2026
Spatially Grounded Concept Bottleneck Models via Part-Factorized Attention
arXiv · 5 iun. 2026
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · 18 sept. 2026