Position: Explanation Stability Is a Property of the Model Method Pair, Not the Model

2026-07-18

AI Summary

This study challenges the common assumption that explanation stability is an inherent property of deep learning models, demonstrating instead that it emerges from the interaction between the model and the attribution method. Systematically evaluating DenseNet201, ResNet50V2, and InceptionV3 on chest X-ray data using multiple attribution techniques—including LayerCAM and GradCAM++—the authors employ IoU and AUC (both exceeding 99%) as stability metrics. Their results reveal that model stability rankings vary dramatically across attribution methods: for instance, LayerCAM identifies InceptionV3 as the most stable model (IoU = 0.777), whereas GradCAM++ reduces its stability by 17.3%. These findings underscore that attribution method choice critically influences stability assessments and highlight the necessity of cross-method validation to avoid biased conclusions.

Abstract

This position paper argues that claims about explanation stability are scientifically invalid without cross method validation. Just as statistical significance requires the test statistic to be specified, stability should either be evaluated across multiple attribution paradigms or explicitly scoped to the computational objective of a single method. In controlled chest X ray experiments, DenseNet201, ResNet50V2, and InceptionV3 achieved AUC values above 99%, yet their stability rankings reversed across attribution methods. LayerCAM ranked InceptionV3 as the most stable model, with an IoU of 0.777, whereas GradCAM++ favored DenseNet201 and reduced InceptionV3 stability score by 17.3%. These findings demonstrate that explanation stability is an emergent property of the model method pair rather than an intrinsic characteristic of the model alone. We therefore argue that explanation based claims should be validated across multiple attribution methods and that regulatory submissions should explicitly specify the attribution operators used to avoid creating illusory safety assurances.

Problem

• explanation stability
• attribution methods
• model interpretability
• cross-method validation
• regulatory safety

Innovation

• explanation stability
• attribution methods
• model-method pair
• cross-method validation
• interpretability