A Generic Framework for Black-box Explanations

Abstract

Beyond their differences, most black-box explanation methods share a number of features and can be framed in a common structure. We identify these features and propose a generic and parameterized framework which makes it possible to combine them in different ways. This framework has been implemented in a proof of concept system called IBEX (for "Interactive Black-box EXplanation system"). IBEX makes it possible to address a variety of needs of different types of explainees (e.g. local or global explanations, detailed or simple explanations, explanations in the form of counterfactuals, rules, plots, etc.). We illustrate the benefit of the approach in terms of versatility through several case studies corresponding to different types of explainees.

Confier une décision vitale à une machine

“…À la suite notamment d'un appel de Tim Miller en 2017 (Miller et al, 2017), l'intérêt des sciences sociales pour la conception de ces explications est reconnu par une partie de la communauté (Miller, 2017 ;Mueller et al, 2019), notamment pour identifier les mécanismes à l'oeuvre dans les processus d'explication humains. Cela a par exemple conduit au développement de méthodes d'explication interactives (Henin et Le Métayer, 2020b ;2021) et à des méthodes de justifications d'algorithmes (Henin et Le Métayer, 2020a).…”

One Explanation Does Not Fit All: The Promise of Interactive Explanations for Machine Learning Transparency

“…Miller et al [27] noticed that explainability systems built for autonomous agents and predictive systems rarely ever consider the end users and their expectations as they are mostly "built by engineers, for engineers." Since then, XAI and IML research has taken a more human-centred direction, with many academics and engineers [40,39,34,44,12] evaluating their approaches against Miller's guidelines to help mitigate such issues.…”

“…The latter approach does not, however, require the explainability system to be interactive as the same personalisation can be achieved off-line by extracting the personalisation specification from the explainee and subsequently incorporating it into the data or algorithm initialisation. Interaction with explainability systems has also been acknowledged by Henin and Le Métayer [12], who proposed a generic mathematical formulation of black-box explainers consisting of three distinct steps: sampling, generation and interaction.…”

A Performance-Explainability Framework to Benchmark Machine Learning Methods: Application to Multivariate Time Series Classifiers

“…As our framework aims to cover all types of methods, we do not present the frameworks focusing on a particular type of explainability methods (e.g. [Ancona et al, 2018;Henin et al, 2019;Lundberg et al, 2017]).…”