The article discusses the Model Parameter Randomisation Test (MPRT), a widely acknowledged method in the eXplainable Artificial Intelligence (XAI) community. However, recent studies have identified issues with the empirical interpretation of MPRT. To address these, the authors propose two adaptations: SmoothMPRT and EfficientMPRT. The former minimizes the impact of noise on evaluation results, while the latter circumvents the need for biased similarity measurements. The experimental results showed that these adaptations lead to improved metric reliability, thus enabling a more trustworthy application of XAI methods.
Publication date: 12 Jan 2024
Project Page: https://arxiv.org/abs/2401.06465v1
Paper: https://arxiv.org/pdf/2401.06465
