Computer Vision & Multimodal AI · arXiv:2610.35827
Seeing Through Distractions: Context-Stable Attribution via the Least Core
Authors: Sai Ganesh Nagarajan, Toshinori Yamauchi, Hiroshi Kera (arXiv:2610.35827) · arXiv: 2610.35827
Core Methodological Innovation
Featured in Nujan's Computer Vision & Multimodal AI Literature Digest. Open side-by-side with interactive equation, methodology, and citation breakdowns.
Key Quantitative & Theoretical Takeaway: Open arXiv:2610.35827 in Nujan to inspect every mathematical symbol, experimental baseline, and citation directly alongside the original PDF.
Abstract
Shapley-value-based explanations are widely used for feature attribution in machine learning. However, in computer vision, a feature may reduce the target-class score when added to the complete genuine image, even when the predicted label remains unchanged. We formalize such features as contextual distractors and investigate their attributions under semivalues and the least core. Under a weak average marginal monotonicity assumption, we derive quantitative semivalue--least-core gap bounds and sufficient conditions based on the probability of sampling smaller contexts. We further show that shifting semivalue weights toward larger contexts decreases the signed gap. Finally, controlled image-classification experiments and evaluations on Salient ImageNet provide strong empirical support for our theoretical findings.
Step-by-Step Equation & Methodology Breakdown
What is Seeing Through Distractions: Context-Stable Attribution via the Least Core (arXiv:2610.35827) about?
Shapley-value-based explanations are widely used for feature attribution in machine learning. However, in computer vision, a feature may reduce the target-class score when added to the complete genuine image, even when the predicted label remains unchanged. We formalize such features as contextual distractors and investigate their attributions under semivalues and the least core. Under a weak average marginal monotonicity assumption, we derive quantitative semivalue--least-core gap bounds and sufficient conditions based on the probability of sampling smaller contexts. We further show that shifting semivalue weights toward larger contexts decreases the signed gap. Finally, controlled image-classification experiments and evaluations on Salient ImageNet provide strong empirical support for our theoretical findings.
How can I read arXiv:2610.35827 with interactive explanations?
Visit https://nujan.app/abs/2610.35827 to open Seeing Through Distractions: Context-Stable Attribution via the Least Core in Nujan's split-screen research workspace with instant equation and methodology breakdowns.