minor updates to abstract

paper/paper.tex

@@ -106,7 +106,7 @@ Energy-Constrained Conformal Counterfactuals}
 
 
 \begin{abstract}
-  Counterfactual Explanations offer an intuitive and straightforward way to explain black-box models but they are not unique. To identify the most plausible explanations, existing work has primarily relied on surrogate models to learn how the input data is distributed. This effectively reallocates the task of learning realistic representations of the data from the model itself to the surrogate. Consequently, the generated explanations may look plausible to humans but not necessarily faithfully describe the behaviour of the black-box model. We formalise this notion of faithfulness through the introduction of a tailored evaluation metric and propose a novel algorithmic framework for generating \textbf{E}nergy-\textbf{C}onstrained \textbf{C}onformal \textbf{Co}unterfactuals that are only as plausible as the model permits. Through extensive empirical studies involving multiple synthetic and real-world datasets, we demonstrate that ECCCos reconcile the need for plausibility and faithfulness. In particular, we show that it is possible to achieve state-of-the-art plausibility for any black-box model with gradient access without the need for surrogate models. To do so, \textbf{ECCCo} relies solely on properties defining the black-box model itself by leveraging recent advances in energy-based modelling and conformal inference. The empirical findings also highlight that black-box models that are trained jointly to discriminate outputs and generate inputs tend to yield more plausible explanations than pure discriminative models. Our framework is intuitive, flexible and open-sourced. By highlighting the need for faithfulness in the context of Counterfactual Explanations, we believe that in the short term, our work will enable researchers and practitioners to better distinguish trustworthy from unreliable models. We anticipate that ECCCo can serve as a baseline for future research directed at providing plausible but faithful Counterfactual Explanations. 
+  Counterfactual Explanations offer an intuitive and straightforward way to explain black-box models and offer Algorithmic Recourse to individuals. To address the need for plausible explanations, existing work has primarily relied on surrogate models to learn how the input data is distributed. This effectively reallocates the task of learning realistic representations of the data from the model itself to the surrogate. Consequently, the generated explanations may seem plausible to humans but need not necessarily faithfully describe the behaviour of the black-box model. We formalise this notion of faithfulness through the introduction of a tailored evaluation metric and propose a novel algorithmic framework for generating \textbf{E}nergy-\textbf{C}onstrained \textbf{C}onformal \textbf{Co}unterfactuals that are only as plausible as the model permits. Through extensive empirical studies involving multiple synthetic and real-world datasets, we demonstrate that \textbf{ECCCo} reconciles the need for plausibility and faithfulness. In particular, we show that it is possible to achieve state-of-the-art plausibility for models with gradient access without the need for surrogate models. To do so, ECCCo relies solely on properties defining the black-box model itself by leveraging recent advances in energy-based modelling and conformal inference. Through this work, we also shine new light on the explanatory properties of Joint Energy Models. Our framework is intuitive, flexible and fully open-sourced. By highlighting the need for faithfulness in the context of Counterfactual Explanations, we believe that in the short term, our work will enable researchers and practitioners to better distinguish trustworthy from unreliable models. We further anticipate that ECCCo can serve as a baseline for future research directed at providing plausible but faithful Counterfactual Explanations. 
 \end{abstract}
 
 \section{Introduction}\label{intro}