Comparison-Based Inverse Classification for Interpretability in Machine Learning

By Thibault Laugel, Marie-Jeanne Lesot, C. Marsala, Xavier Renard, Marcin Detyniecki. 2018

133 citations recorded by the source corpus.

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

The Dangers of Post-hoc Interpretability: Unjustified Counterfactual Explanations (2019, methodology): “For instance, [ Laugel et al. , 2018a ] (resp.”
Counterfactual Explanations for Machine Learning: A Review (2020, methodology): “Some of the algorithms cannot handle categorical variables and ﬁlter them out [74, 79].”
A Survey of Contrastive and Counterfactual Explanation Generation Methods for Explainable Artificial Intelligence (2021, methodology): “, where the same method is used to output images for a handwritten digit classification problem [143].”
Interpretable Counterfactual Explanations Guided by Prototypes (2019, methodology): “One of the drawbacks of this and other techniques for finding counterfactual instances (Wachter, Mittelstadt, and Russell 2018; Laugel et al. 2018; Dhurandhar et al. 2018) is that they don’t work out of the box for categorical variables.”
RAPID: Rapid and Precise Interpretable Decision Sets (2019, background): “Some studies focus on generating counterfactual explanations [13], [14], where the goal is to identify minimal changes required to change the model’s prediction.”
Defining Locality for Surrogates in Post-hoc Interpretablity (2018, methodology): “This is done by using the Generation part of the GrowingSpheres algorithm introduced in (Laugel et al., 2018).”
Explaining Deep Learning Models with Constrained Adversarial Examples (2019, background): “[7] finds synthetic counterfactual explanations by sampling from a sphere around the input in a growing radius, until one is found which classifies differently than the original.”
Data-centric Artificial Intelligence: A Survey (2023, unknown): “White-box methods necessitate access to the evaluated models, which can be achieved through complete internals [36, 111, 138], gradients [225], or solely the prediction function [52, 57, 127, 202].”