Explaining explanations: axiomatic feature interactions for deep networks

MaRDI QIDQ4999041zbMATH OpenFDO

Authors Joseph D. Janizek, Pascal Sturmfels, Su-In Lee

Publication date 9 July 2021

Full work available at URL https://arxiv.org/abs/2002.04138, https://jmlr.csail.mit.edu/papers/v22/20-1223.html

zbMATH Keywords

Aumann-Shapley valuefeature attributionfeature interactioninterpretabilityneural networks

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05)

Recommendations

• Interpreting deep learning models with marginal attribution by conditioning on quantiles
• Towards formal XAI: formally approximate minimal explanations of neural networks
• Explainable deep learning: a field guide for the uninitiated
• Explaining individual predictions when features are dependent: more accurate approximations to Shapley values

Cites work

• An axiomatic approach to the concept of interaction among players in cooperative games
• An efficient explanation of individual classifications using game theory
• Augmentable gamma belief networks
• Axiomatic characterizations of generalized values
• Evaluating groups with the generalized Shapley value
• On the computational complexity of weighted voting games

Cited in

(5)

• PredDiff: explanations and interactions from conditional expectations
• Mutual information for explainable deep learning of multiscale systems
• Building the hierarchical Choquet integral as an explainable AI classifier via neuroevolution and pruning
• Towards formal XAI: formally approximate minimal explanations of neural networks
• Four axiomatic characterizations of the integrated gradients attribution method

Describes a project that uses

Uses Software

• TensorFlow
• PyTorch
• DrugBank
• Transformers
• DistilBERT
• shap

This page was built for publication: Explaining explanations: axiomatic feature interactions for deep networks

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4999041)