Maximizing Invariant Data Perturbation with Stochastic Optimization

TL;DR

This paper introduces a stochastic optimization method for feature attribution in deep neural networks, reformulating the complex perturbation problem into a differentiable maximization task solvable by gradient-based algorithms, demonstrated on image classification.

Contribution

It presents a novel reformulation of perturbation-based feature attribution as a differentiable maximization problem solvable by stochastic gradient methods.

Findings

Effective identification of relevant image regions.

Compatibility with popular gradient-based optimizers.

Improved interpretability of neural network decisions.

Abstract

Feature attribution methods, or saliency maps, are one of the most popular approaches for explaining the decisions of complex machine learning models such as deep neural networks. In this study, we propose a stochastic optimization approach for the perturbation-based feature attribution method. While the original optimization problem of the perturbation-based feature attribution is difficult to solve because of the complex constraints, we propose to reformulate the problem as the maximization of a differentiable function, which can be solved using gradient-based algorithms. In particular, stochastic optimization is well-suited for the proposed reformulation, and we can solve the problem using popular algorithms such as SGD, RMSProp, and Adam. The experiment on the image classification with VGG16 shows that the proposed method could identify relevant parts of the images effectively.