Software for Dataset-wide XAI: From Local Explanations to Global\n Insights with Zennit, CoRelAy, and ViRelAy

Abstract

Deep Neural Networks (DNNs) are known to be strong predictors, but their\nprediction strategies can rarely be understood. With recent advances in\nExplainable Artificial Intelligence (XAI), approaches are available to explore\nthe reasoning behind those complex models' predictions. Among post-hoc\nattribution methods, Layer-wise Relevance Propagation (LRP) shows high\nperformance. For deeper quantitative analysis, manual approaches exist, but\nwithout the right tools they are unnecessarily labor intensive. In this\nsoftware paper, we introduce three software packages targeted at scientists to\nexplore model reasoning using attribution approaches and beyond: (1) Zennit - a\nhighly customizable and intuitive attribution framework implementing LRP and\nrelated approaches in PyTorch, (2) CoRelAy - a framework to easily and quickly\nconstruct quantitative analysis pipelines for dataset-wide analyses of\nexplanations, and (3) ViRelAy - a web-application to interactively explore\ndata, attributions, and analysis results. With this, we provide a standardized\nimplementation solution for XAI, to contribute towards more reproducibility in\nour field.\n

Ähnliche Arbeiten