New PEAR method boosts multimodal reasoning in vision-language models

Researchers have developed PEAR, an adaptive method to improve multimodal reasoning in large vision-language models. PEAR enhances the model's ability to perceive visual information and explore diverse reasoning paths by adaptively reshaping sequence-level advantages. This method adds an entropy-based exploration gain specifically at high-entropy positions, with lower visual dependency resulting in a larger gain. Experiments on Qwen2.5-VL-3B and Qwen2.5-VL-7B models demonstrated that PEAR outperforms existing baselines on multimodal reasoning benchmarks, as well as textual reasoning and general multimodal tasks. AI

IMPACT This new method could enhance the capabilities of large vision-language models in complex reasoning tasks.

RANK_REASON The cluster contains a research paper detailing a new method for improving AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Hugging Face
• Qwen2.5-VL-3B
• Qwen2.5-VL-7B

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →