Disentangling Perception and Reasoning in Multimodal LLMs via Reward Design
cs.CV, cs.CL
Submitted: 2026-01-01
Updated: 2026-10-31
Code: https://github.com/hiyouga/EasyR1
Terminology
Sources
- Diagnosing Visual Reasoning: Challenges, Insights, and a Path Forward
- Ground-R1: Incentivizing Grounded Visual Reasoning via Reinforcement Learning
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- GRIT: Teaching MLLMs to Think with Images
- Have Multimodal Large Language Models (MLLMs) Really Learned to Tell the Time on Analog Clocks?
- VisioMath: Benchmarking Figure-based Mathematical Reasoning in LMMs
- Training Vision-Language Process Reward Models for Test-Time Scaling in Multimodal Reasoning: Key Insights and Lessons Learned
- Spurious Rewards: Rethinking Training Signals in RLVR
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model
- The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
- VisualPuzzles: Decoupling Multimodal Reasoning Evaluation from Domain Knowledge
- Thinking with Images for Multimodal Reasoning: Foundations, Methods, and Future Frontiers
- The Jumping Reasoning Curve? Tracking the Evolution of Reasoning Performance in GPT-[n] and o-[n] Models on Multimodal Puzzles
- VisualPRM: An Effective Process Reward Model for Multimodal Reasoning
- Qwen3 Technical Report
- Qwen2.5 Technical Report
- R1-Onevision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalization
- DAPO: An Open-Source LLM Reinforcement Learning System at Scale
- How Far Are We from Intelligent Visual Deductive Reasoning?
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs