Geometric Remove-and-Retrain (GOAR): Coordinate-Invariant eXplainable AI Assessment

Published: 27 Oct 2023, Last Modified: 23 Nov 2023NeurIPS XAIA 2023EveryoneRevisionsBibTeX
TL;DR: We illuminate the pitfalls of Remove-and-Retrain (ROAR) from a geometrical perspective and propose a new evaluation metric called Geometrical Remove-and-Retrain (GOAR) that can resolve the issues of ROAR.
Abstract: Identifying the relevant input features that have a critical influence on the output results is indispensable for the development of explainable artificial intelligence (XAI). Remove-and-Retrain (ROAR) is a widely accepted approach for assessing the importance of individual pixels by measuring changes in accuracy following their removal and subsequent retraining of the modified dataset. However, we uncover notable limitations in pixel-perturbation strategies. When viewed from a geometric perspective, this method perturbs pixels by moving each sample in the pixel-basis direction. However, we have found that this approach is coordinate-dependent and fails to discriminate between differences among features, thereby compromising the reliability of the evaluation. To address this challenge, we introduce an alternative feature-perturbation approach named Geometric Remove-and-Retrain (GOAR). GOAR offers a perturbation strategy that takes into account the geometric structure of the dataset, providing a coordinate-independent metric for accurate feature comparison. Through a series of experiments with both synthetic and real datasets, we substantiate that GOAR's geometric metric transcends the limitations of pixel-centric metrics.
Submission Track: Full Paper Track
Application Domain: Computer Vision
Survey Question 1: We shed light on the limitations of Remove-and-Retrain (ROAR), one of the influential feature attribution evaluation metrics, from a geometrical perspective. Additionally, we propose a new metric called Geometrical Remove-and-Retrain (GOAR) that can address these limitations.
Survey Question 2: In the field of Explainable AI (XAI), feature attribution methods are of utmost importance. Therefore, research on metrics to evaluate them is also crucial. Our geometrical perspective could provide novel insights into feature attribution methods and their evaluation benchmarks.
Survey Question 3: In our research, we utilized input gradient, input times gradient, SmoothGrad, and integrated gradient (IG).
Submission Number: 1