Interactive eXplainable Artificial Intelligence (IXAI)

Description:

The IXAI toolbox enhances AI model reliability by providing clear decision-making insights. It allows domain experts to review and correct explanations, improving model performance and trustworthiness. Validated in epilepsy detection, IXAI requires minimal annotations and meets European AI Act requirements. It includes five explanation methods and boosts Imagenet dataset accuracy by 5%.

More Information:

• Enrichment project: Interactive Explainable AI

• Responsible Research Lead: Maarten De Vos (KU Leuven)
  

• Authors: Joran Michiels
  

• Programming Language: Python

• Link to the repository: GitHub - JoranMichiels/ixai: Interactive eXplainable Artificial Intelligence (IXAI), enforce your Pytorch model to have correct local post-hoc explanations 

• Categories: Explainability, Model Correction, Human Oversight
  

Publications:

5 min pitch on Interactive eXplainable Artificial Intelligence