Loss Input Susceptibility Score

Loss Input Susceptibility Score (LISS), or related metrics like susceptibility, quantifies a system's sensitivity to perturbations or noise, with applications ranging from assessing the robustness of machine learning models to characterizing individual vulnerability to misinformation. Current research focuses on developing computational models to estimate susceptibility, often leveraging gradient information or unlabeled data, and employing architectures like U-Nets for image processing tasks. Understanding and mitigating susceptibility is crucial for improving the reliability of AI systems and for developing strategies to combat the spread of misinformation.