Reproducibility in XAI Research: Open Source Benchmarks for Explanation Quality

Key Points

• The aim is to address reproducibility issues in explainable artificial intelligence (XAI) research by providing benchmarks for explanation quality.
• Developed open-source benchmarks for evaluating explanation quality in XAI systems.
• Assessed the reproducibility of XAI results using the established benchmarks.
• Identified significant variability in the explanation quality across different XAI models.
• Demonstrated improved reproducibility of XAI outputs when using standardized benchmarks.