Local Interpretable Model-agnostic Explanations of Bayesian Predictive Models via Kullback-Leibler Projections

Abstract

We introduce a method, KL-LIME, for explaining predictions of Bayesian predictive models by projecting the information in the predictive distribution locally to a simpler, interpretable explanation model. The proposed approach combines the recent Local Interpretable Model-agnostic Explanations (LIME) method with ideas from Bayesian projection predictive variable selection methods. The information theoretic basis helps in navigating the trade-off between explanation fidelity and complexity. We demonstrate the method in explaining MNIST digit classifications made by a Bayesian deep convolutional neural network.

arXiv commentExtended abstract/short paper, Proceedings of the 2nd Workshop on Explainable Artificial Intelligence (XAI 2018) at IJCAI/ECAI 2018

References in corpus (2)

• Bayesian Convolutional Neural Networks with Bernoulli Approximate Variational Inference
• Projection predictive model selection for Gaussian processes

Cited by in corpus (6)

• Opportunities and Challenges in Explainable Artificial Intelligence (XAI): A Survey
• Projective Inference in High-dimensional Problems: Prediction and Feature Selection
• AI for Explaining Decisions in Multi-Agent Environments
• Why model why? Assessing the strengths and limitations of LIME
• LIMEcraft: Handcrafted superpixel selection and inspection for Visual eXplanations
• ProtoShotXAI: Using Prototypical Few-Shot Architecture for Explainable AI