Lewis: Explaining Black-Box Algorithms Using Probabilistic Contrastive Counterfactuals
Home Paper Contact Us


Overview

Lewis is a causality-based system that uses probabilistic contrastive counterfactuals for generating post-hoc explanations for black-box decision-making algorithms. First, it provides insights into what causes an algorithm's decisions at the global, local and contextual (sub-population) levels. The explanations generated by Lewis are described in terms of three novel probabilistic measures-- necessity score, sufficiency score and necessity and sufficiency score-- that quantify the influence of attributes toward an algorithm's decision. Furthermore, for individuals negatively impacted by the algorithm's decisions, Lewis generates actionable recourse translatable into real-world interventions.

Short summary



Papers:

Explaining Black-Box Algorithms Using Probabilistic Contrastive Counterfactuals.
Sainyam Galhotra*, Romila Pradhan* and Babak Salimi. SIGMOD 2021.

Demonstration of Generating Explanations for Black-BoxAlgorithms UsingLewis.
Paul Y. Wang, Sainyam Galhotra*, Romila Pradhan* and Babak Salimi. VLDB 2021.

20min talk



Demo Video

Contributors: Paul Y. Wang, Sainyam Galhotra, Romila Pradhan, and Babak Salimi

Please reach out to any of the contributors if you have any questions.