Attribution Mask: Filtering Out Irrelevant Features By Recursively Focusing Attention on Inputs of DNNs

TL;DR

This paper introduces an attribution mask method that filters irrelevant input features to improve DNN classification accuracy, supported by theoretical insights and a new attribution technique called GxSI, achieving near-perfect accuracy on CIFAR-10.

Contribution

The paper proposes a recursive attention-based attribution mask that enhances DNN accuracy by filtering irrelevant features, with a novel GxSI method for mask generation and theoretical validation.

Findings

Accuracy improved to 99.8-99.9% on CIFAR-10 without retraining.

The method is effective under the no implicit bias condition.

Theoretical analysis links the approach to compressing DNNs into single-layer networks.

Abstract

Attribution methods calculate attributions that visually explain the predictions of deep neural networks (DNNs) by highlighting important parts of the input features. In particular, gradient-based attribution (GBA) methods are widely used because they can be easily implemented through automatic differentiation. In this study, we use the attributions that filter out irrelevant parts of the input features and then verify the effectiveness of this approach by measuring the classification accuracy of a pre-trained DNN. This is achieved by calculating and applying an \textit{attribution mask} to the input features and subsequently introducing the masked features to the DNN, for which the mask is designed to recursively focus attention on the parts of the input related to the target label. The accuracy is enhanced under a certain condition, i.e., \textit{no implicit bias}, which can be derived based on our theoretical insight into compressing the DNN into a single-layer neural network. We also provide Gradient\,*\,Sign-of-Input (GxSI) to obtain the attribution mask that further improves the accuracy. As an example, on CIFAR-10 that is modified using the attribution mask obtained from GxSI, we achieve the accuracy ranging from 99.8\% to 99.9\% without additional training.