Research1h ago
Attention Maps Fail To Predict VLM Accuracy
A study of LLaVA-1.5, PaliGemma, and Qwen2-VL shows that attention structure is a near-zero predictor of correctness. Researchers used a new VLM Reliability Probe to debunk the intuition that sharp attention maps imply calibrated answers. This suggests that visual confidence is not stored in attention weights, complicating efforts to detect model hallucinations.