Attribution-Scores in Data Management and Explainable Machine Learning

TL;DR

This paper explores the use of actual causality to define responsibility scores as explanations for database query answers and machine learning model outcomes, linking to database repairs and model interpretability.

Contribution

It introduces a unified approach to responsibility scores using actual causality, connecting database repairs with explanation methods in machine learning.

Findings

Responsibility scores can be used to explain query answers and model outcomes.

Database repairs help measure database consistency and support responsibility calculations.

Efficient computation methods for Shap-score are analyzed.

Abstract

We describe recent research on the use of actual causality in the definition of responsibility scores as explanations for query answers in databases, and for outcomes from classification models in machine learning. In the case of databases, useful connections with database repairs are illustrated and exploited. Repairs are also used to give a quantitative measure of the consistency of a database. For classification models, the responsibility score is properly extended and illustrated. The efficient computation of Shap-score is also analyzed and discussed. The emphasis is placed on work done by the author and collaborators.