A Fresh Look at Sanity Checks for Saliency Maps
This work tackles incremental improvements in evaluation methods for explainable AI, benefiting researchers and practitioners in the XAI community.
The paper addresses methodological concerns in the Model Parameter Randomisation Test (MPRT) for saliency maps by proposing Smooth MPRT and Efficient MPRT, which improve metric reliability for more trustworthy explanation methods.
The Model Parameter Randomisation Test (MPRT) is highly recognised in the eXplainable Artificial Intelligence (XAI) community due to its fundamental evaluative criterion: explanations should be sensitive to the parameters of the model they seek to explain. However, recent studies have raised several methodological concerns for the empirical interpretation of MPRT. In response, we propose two modifications to the original test: Smooth MPRT and Efficient MPRT. The former reduces the impact of noise on evaluation outcomes via sampling, while the latter avoids the need for biased similarity measurements by re-interpreting the test through the increase in explanation complexity after full model randomisation. Our experiments show that these modifications enhance the metric reliability, facilitating a more trustworthy deployment of explanation methods.