Visual Attention Reasoning via Hierarchical Search and Self-Verification

TL;DR

This paper introduces Visual Attention Reasoning (VAR), a reinforcement learning framework that improves multimodal large language models by enabling hierarchical search and self-verification to reduce hallucinations and enhance visual grounding.

Contribution

It presents a novel hierarchical search and self-verification framework with explicit visual grounding, backed by theoretical validation and superior experimental performance.

Findings

Significantly reduces hallucinations in MLLMs

Enforces traceable evidence grounding with bounding boxes

Outperforms state-of-the-art methods on safety benchmarks

Abstract

Multimodal Large Language Models (MLLMs) frequently hallucinate due to their reliance on fragile, linear reasoning and weak visual grounding. We propose Visual Attention Reasoning (VAR), a reinforcement learning framework that reformulates reasoning as a hierarchical search with self-verification. VAR enforces traceable evidence grounding by generating explicit bounding boxes, guided by a novel reward function combining geometric precision and semantic sufficiency. Furthermore, it replaces linear Chain-of-Thought with a tree-search policy capable of backtracking to correct logical errors. Theoretical analysis validates the framework's reliability, and extensive experiments demonstrate that VAR significantly outperforms state-of-the-art methods on complex hallucination and safety benchmarks.