Visual explanation of black-box model: Similarity Difference and Uniqueness (SIDU) method

TL;DR

The paper introduces SIDU, a novel visual explanation method for black-box models that effectively localizes object regions responsible for predictions, validated through multiple experiments and robustness tests.

Contribution

The paper presents SIDU, a new visual explanation algorithm that improves localization of relevant regions in black-box models, with comprehensive evaluation and robustness analysis.

Findings

SIDU effectively localizes object regions responsible for predictions.

SIDU outperforms existing methods in various evaluation metrics.

SIDU maintains robustness under adversarial attacks.

Abstract

Explainable Artificial Intelligence (XAI) has in recent years become a well-suited framework to generate human understandable explanations of "black-box" models. In this paper, a novel XAI visual explanation algorithm known as the Similarity Difference and Uniqueness (SIDU) method that can effectively localize entire object regions responsible for prediction is presented in full detail. The SIDU algorithm robustness and effectiveness is analyzed through various computational and human subject experiments. In particular, the SIDU algorithm is assessed using three different types of evaluations (Application, Human and Functionally-Grounded) to demonstrate its superior performance. The robustness of SIDU is further studied in the presence of adversarial attack on "black-box" models to better understand its performance. Our code is available at: https://github.com/satyamahesh84/SIDU_XAI_CODE.