On the Evaluation Consistency of Attribution-based Explanations

TL;DR

This paper introduces Meta-Rank, a benchmarking platform for attribution methods in image classification, revealing inconsistencies in evaluation practices and emphasizing the need for more rigorous, comprehensive assessments.

Contribution

It presents Meta-Rank, an open platform for systematic benchmarking of attribution methods across multiple models and datasets, highlighting evaluation inconsistencies and guiding future research.

Findings

Evaluation results vary significantly under different settings.

Performance rankings are consistent across training checkpoints.

Current evaluation methods do not outperform simple baselines on diverse models.

Abstract

Attribution-based explanations are garnering increasing attention recently and have emerged as the predominant approach towards \textit{eXplanable Artificial Intelligence}~(XAI). However, the absence of consistent configurations and systematic investigations in prior literature impedes comprehensive evaluations of existing methodologies. In this work, we introduce {Meta-Rank}, an open platform for benchmarking attribution methods in the image domain. Presently, Meta-Rank assesses eight exemplary attribution methods using six renowned model architectures on four diverse datasets, employing both the \textit{Most Relevant First} (MoRF) and \textit{Least Relevant First} (LeRF) evaluation protocols. Through extensive experimentation, our benchmark reveals three insights in attribution evaluation endeavors: 1) evaluating attribution methods under disparate settings can yield divergent performance rankings; 2) although inconsistent across numerous cases, the performance rankings exhibit remarkable consistency across distinct checkpoints along the same training trajectory; 3) prior attempts at consistent evaluation fare no better than baselines when extended to more heterogeneous models and datasets. Our findings underscore the necessity for future research in this domain to conduct rigorous evaluations encompassing a broader range of models and datasets, and to reassess the assumptions underlying the empirical success of different attribution methods. Our code is publicly available at \url{https://github.com/TreeThree-R/Meta-Rank}.