Grad-CAM

Findings

• IC-044Tulu-2-13B's internal activations contain a linearly decodable, faithful representation of input-context propositions that persists under prompt injection and backdoor attacks where outputs become unfaithful [supporting]
• IC-1156The VPT advantage over FT for ViT-B/16 is not explained by overfitting resistance or additional optimization dimensions; the specific feature-preservation mechanism of VPT is the key factor [supporting]
• IC-447CLIP ViT-B/16 produces noisy saliency maps and contains only 42 concept detectors, indicating poor visual interpretability [primary]
• IC-449CLIP ViT-B/16 Grad-CAM explanations are highly sensitive to input noise, with SSIM dropping from 91.18% to 70.58% as noise standard deviation increases from 1/255 to 9/255 [primary]
• IC-644CLIP relies on spurious features (background) as a shortcut in zero-shot classification, and conditioning on the correct background reduces this reliance [primary]
• IC-680CLIP ViT's image token contributions are spatially localized to match described content, enabling zero-shot segmentation that outperforms existing CLIP-based methods [compared-to]
• IC-742ResNet-50-BN on Waterbirds relies on background as a spurious feature for classification, and this shortcut is invisible to entropy-based confidence metrics [primary]