A Symbolic Approach to Explaining Bayesian Network Classifiers

By Andy Shih, Arthur Choi, Adnan Darwiche. 2018

291 citations recorded by the source corpus.

We propose an approach for explaining Bayesian network classifiers, which is based on compiling such classifiers into decision functions that have a tractable and symbolic form. We introduce two types of explanations for why a classifier may have classified an instance positively or negatively and suggest algorithms for computing these explanations. The first type of explanation identifies a minimal set of the currently active features that is responsible for the current classification, while the second type of explanation identifies a minimal set of features whose current state (active or not) is sufficient for the classification. We consider in particular the compilation of Naive and Latent-Tree Bayesian network classifiers into Ordered Decision Diagrams (ODDs), providing a context for evaluating our proposal using case studies and experiments based on classifiers from the literature.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Computing Optimal Decision Sets with SAT (2020, background): “Hence the rising interest in explainable artificial intelligence (XAI) [4, 14, 16, 18, 22, 23, 26, 29, 30, 36–38, 42, 43, 48, 49, 52], including research programs [3, 24] and legislation [17, 21].”
On Relating Explanations and Adversarial Examples (2019, methodology): “Given some target prediction π ∈ K, one can devise representations for the formula FM,π , (M→π) [52, 23].”
On Explaining Decision Trees (2020, methodology): “3 A PI-explanation is a subset-minimal set of feature-value pairs that entails the prediction [61].”
Towards Trustable Explainable AI (2020, methodology): “Recent work [ Shih et al. , 2018; Ignatiev et al. , 2019a ] proposed a principled approach to computing provably correct explanations (at the expected cost of lower scalability), which is fundamentally different from the heuristic methods.”
Three Modern Roles for Logic in AI (2020, methodology): “This compilation algorithm was recently extended to Bayesian network classifiers with tree structures [82] and later to Bayesian network classifiers with arbitrary structures [83].”
Abduction-Based Explanations for Machine Learning Models (2018, methodology): “It should also be noted that the proposed approach is constraint-agnostic and computes explanations “of the ﬂy” in the online manner while the work of (Shih, Choi, and Darwiche 2018) relies on prior compilation of the classi-ﬁer function into a BDD, which is in general known to be computationally…”