Countdown to NeurIPS 2019 continues... (5th of 8 studies my team will present) A lot of companies ranging from small startups to large corporate giants are releasing Explainable AI toolkits and core features using popular XAI methods like LIME, SHAP, Integrated Gradients, etc. However, one begs the question: Do the explanations provided by these XAI methods really reflect the decisions made by machine learning algorithm? In this study, we introduce a measurable way in an attempt to answer this question, and study some of the most popular XAI methods to see where they stand for deep neural networks. The results may surprise you...
#deeplearning #neurips
https://arxiv.org/abs/1910.07387

❇️ @AI_Python_EN