Built independently by an author, for readers. Read the story and support ChapterPal
ChapterPal Logo
Log inSign up

keyword

softmax score suppression

Softmax score suppression is the reduction in a neural network classifier's predicted probability for a target class when specific features or regions of an input are altered, masked, or removed. In machine learning interpretability and saliency analysis, this effect serves as a metric to evaluate feature importance by measuring the degree to which a model loses confidence in its decision following deliberate perturbations, such as blurring or occlusion. By determining the smallest input regions whose disruption leads to substantial suppression of the target output score, researchers and practitioners can identify and explain the critical evidence the model relies upon to make its classifications.

1 item

Interpretable Explanations of Black Boxes by Meaningful Perturbation

Interpretable Explanations of Black Boxes by Meaningful Perturbation

Ruth Fong, Andrea Vedaldi

OrganizationsUniversity of Oxford

Why you should read this

Develops a model-agnostic framework that produces faithful visual explanations for black-box image classifiers by optimizing meaningful and testable input perturbations.

As machine learning algorithms are increasingly applied to high impact yet high risk tasks, such as medical diagnosis or autonomous driving, it is critical that researchers can explain how such algorithms arrived at their predictions. In recent years, a number of image saliency methods have been developed to summarize where highly complex neural networks "look" in an image for evidence for their predictions. However, these techniques are limited by their heuristic nature and architectural constraints. In this paper, we make two main contributions: First, we propose a general framework for learning different kinds of explanations for any black box algorithm. Second, we specialise the framework to find the part of an image most responsible for a classifier decision. Unlike previous works, our method is model-agnostic and testable because it is grounded in explicit and interpretable image perturbations.

Added

2026-09-24