On The Reasons Behind Decisions
arXiv:2002.09284
Abstract
Recent work has shown that some common machine learning classifiers can be compiled into Boolean circuits that have the same input-output behavior. We present a theory for unveiling the reasons behind the decisions made by Boolean classifiers and study some of its theoretical and practical implications. We define notions such as sufficient, necessary and complete reasons behind decisions, in addition to classifier and decision bias. We show how these notions can be used to evaluate counterfactual statements such as "a decision will stick even if ... because ... ." We present efficient algorithms for computing these notions, which are based on new advances on tractable Boolean circuits, and illustrate them using a case study.
arXiv commentTo appear in the proceedings of European Conference on Artificial Intelligence (ECAI), Spain, 2020
Cited by in corpus (18)
- On Symbolically Encoding the Behavior of Random Forests
- On Relating 'Why?' and 'Why Not?' Explanations
- On Quantifying Literals in Boolean Logic and Its Applications to Explainable AI
- Efficient Explanations for Knowledge Compilation Languages
- On the Complexity of SHAP-Score-Based Explanations: Tractability via Knowledge Compilation and Non-Approximability Results
- Provably efficient, succinct, and precise explanations
- On Tractable Representations of Binary Neural Networks
- Foundations of Symbolic Languages for Model Interpretability
- On the Complexity of Enumerating Prime Implicants from Decision-DNNF Circuits
- Efficient Explanations With Relevant Sets
- On the Explanatory Power of Decision Trees
- On Efficiently Explaining Graph-Based Classifiers
- On Explaining Random Forests with SAT
- Sufficient reasons for classifier decisions in the presence of constraints
- The Tractability of SHAP-Score-Based Explanations over Deterministic and Decomposable Boolean Circuits
- Consistent Sufficient Explanations and Minimal Local Rules for explaining regression and classification models
- Influence-Driven Explanations for Bayesian Network Classifiers
- Explanation from Specification