Thinking with Visual Abstract: Enhancing Multimodal Reasoning via Visual Abstraction Permalink
Published in NeurIPS 2025 Multimodal Algorithmic Reasoning Workshop (ICML Under Review), 2025
This paper proposes Visual Abstract Thinking (VAT), a novel reasoning paradigm inspired by human visual cognition. VAT enables MLLMs to reason more effectively by leveraging visual abstractions rather than explicit verbal Chain-of-Thought.