Where Reliability Lives in Vision-Language Models: A Mechanistic Study of Attention, Hidden States, and Causal Circuits

Logan Mann, Ajit Saravanan, Ishan Dave, Shikhar Shiromani, Saadullah Ismail, Yi Xia, Emily Huang

arXiv:2605.08200 · 2026-05-12 공개 · arXiv · PDF

vision-language-models attention-mechanisms late-fusion early-fusion neuron-ablation layer-wise-margin vlm-architecture hidden-state-analysis

Abstract

A pervasive intuition holds that vision-language models (VLMs) are most trustworthy when their attention maps look sharp: concentrated attention on the queried region should imply a confident, calibrated answer. We test this Attention-Confidence Assumption directly. We instrument three open-weight VLM families (LLaVA-1.5, PaliGemma, Qwen2-VL; 3-7B parameters) with a unified mechanistic pipeline -- the VLM Reliability Probe (VRP) -- that compares attention structure, generation dynamics, and hidden-state geometry against a single correctness label. Three results emerge. (i) Attention structure is a near-zero predictor of correctness (R_pb(C_k,y)=0.001, 95% CI [-0.034,0.036]; R_pb(H_s,y)=-0.012, [-0.047,0.024] on a pooled n=3,090 split), even though attention remains causally necessary for feature extraction (top-30% patch masking drops accuracy by 8.2-11.3 pp, p<0.001). (ii) Reliability becomes legible later in the computation: a single hidden-state linear probe reaches AUROC>0.95 on POPE for two of three families, and self-consistency at K=10 is the strongest behavioral predictor we measure at 10x inference cost (R_pb=0.43). (iii) Causal neuron-level ablations expose a sharp architectural split with direct monitor-design implications: late-fusion LLaVA concentrates reliability in a fragile late bottleneck (-8.3 pp object-identification accuracy after top-5 probe-neuron ablation), whereas early-fusion PaliGemma and Qwen2-VL distribute it widely and absorb destruction of ~50% of their peak-layer hidden dimension with <=1 pp degradation. The takeaway is narrow but consequential: in 3-7B VLMs, reliability is read more reliably off hidden-state geometry, layer-wise margin formation, and sparse late-layer circuits than off attention-map sharpness.

한국어 요약

📋 한 줄 요약

**[VLM Interpretability / Reliability]** VLM의 신뢰성이 어텐션 맵의 선명도가 아닌 후기 층 히든 스테이트 기하, 마진 형성, 희소한 후기 회로에서 더 정확히 읽힌다는 기계적 증거를 제시한다.

🎯 핵심 기여도

💡 핵심 아이디어

VLM이 '어떻게 답하는가'(어텐션 사용)와 '얼마나 신뢰할 수 있는가'(정답 확신도)는 서로 다른 회로에 있다. 신뢰성은 어텐션이 아니라 히든 스테이트 기하와 후기 층의 희소 회로에서 읽혀야 한다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: VLM 신뢰성 모니터링을 어텐션 맵 시각화에서 히든 스테이트 기반 프로빙으로 전환할 근거를 제공한다. **한계**: 평가가 3–7B 파라미터 3개 모델 패밀리와 POPE 등 특정 벤치마크로 제한되며, 더 큰 모델로의 일반화는 추가 연구가 필요하다.

🚀 실용적 활용