Research13h ago
Attention Maps Poorly Predict VLM Reliability
A study of LLaVA-1.5, PaliGemma, and Qwen2-VL reveals that attention structure is a near-zero predictor of correctness. Researchers used the VLM Reliability Probe to debunk the intuition that sharp attention maps signal confident answers. This finding warns practitioners against using attention heatmaps as proxies for model trust or calibration in vision-language tasks.