[Paper Review] Interpretable and Interactive Summaries of Actionable Recourses.

Kaivalya Rawal, Himabindu Lakkaraju|arXiv (Cornell University)|Sep 15, 2020

Explainable Artificial Intelligence (XAI)20 references4 citations

TL;DR

This paper proposes Actionable Recourse Summaries (AReS), a model-agnostic framework that generates global, interpretable, and interactive counterfactual explanations for entire populations. By optimizing for recourse correctness, interpretability, and low overall cost, AReS produces compact rule sets that reveal actionable, non-discriminatory recourses across subpopulations, with theoretical guarantees and empirical validation in real-world datasets.

ABSTRACT

As predictive models are increasingly being deployed in high-stakes decision-making, there has been a lot of interest in developing algorithms which can provide recourses to affected individuals. While developing such tools is important, it is even more critical to analyse and interpret a predictive model, and vet it thoroughly to ensure that the recourses it offers are meaningful and non-discriminatory before it is deployed in the real world. To this end, we propose a novel model agnostic framework called Actionable Recourse Summaries (AReS) to construct global counterfactual explanations which provide an interpretable and accurate summary of recourses for the entire population. We formulate a novel objective which simultaneously optimizes for correctness of the recourses and interpretability of the explanations, while minimizing overall recourse costs across the entire population. More specifically, our objective enables us to learn, with optimality guarantees on recourse correctness, a small number of compact rule sets each of which capture recourses for well defined subpopulations within the data. Our framework is also interactive i.e., it allows users to input specific features of interest which will in turn be used to characterize subpopulations when generating recourse summaries. We also demonstrate theoretically that several of the prior approaches proposed to generate recourses for individuals are special cases of our framework. Experimental evaluation with real world datasets and user studies demonstrate that our framework can provide decision makers with a comprehensive overview of recourses corresponding to any black box model, and consequently help detect undesirable model biases and discrimination.

Motivation & Objective

• To address the critical need for interpretability and fairness vetting in predictive models before real-world deployment, especially in high-stakes decisions.
• To develop a global, model-agnostic framework that summarizes actionable recourses for entire populations rather than individual instances.
• To jointly optimize for correctness of recourses, interpretability of explanations, and minimal overall recourse cost across the population.
• To enable interactive user input for feature-specific subpopulation characterization during recourse summary generation.
• To detect and expose model biases and discriminatory patterns through comprehensive, human-readable summaries of recourses.

Proposed method

• Formulates a novel optimization objective that balances recourse correctness, interpretability of rule sets, and total recourse cost across the population.
• Uses a global, rule-based representation to generate compact, human-readable counterfactual explanations for well-defined subpopulations.
• Employs a constraint-based learning approach to ensure that each rule set provides valid, actionable recourses with optimality guarantees.
• Integrates user-specified features to dynamically define subpopulations for targeted recourse summarization.
• Supports black-box model compatibility by being model-agnostic, enabling application to any trained classifier.
• Theoretically unifies prior individual-level recourse methods as special cases under the proposed framework.

Experimental results

Research questions

• RQ1 Can a global, interpretable summary of recourses be generated for an entire population while preserving correctness and minimizing cost?
• RQ2 How can interpretability and fairness in recourse generation be jointly optimized across diverse subpopulations?
• RQ3 To what extent can user-defined features improve the relevance and specificity of recourse summaries?
• RQ4 How do the proposed rule sets compare to individual-level recourse methods in capturing population-wide patterns?
• RQ5 Can the framework detect and expose hidden model biases and discriminatory patterns through summary-level analysis?

Key findings

• The AReS framework successfully generates global, interpretable, and accurate recourse summaries for entire populations with theoretical optimality guarantees.
• The method produces compact rule sets that capture actionable recourses for well-defined subpopulations, improving transparency and auditability.
• User interaction via feature input enables targeted subpopulation analysis, enhancing relevance and contextual precision of the summaries.
• Experimental evaluation on real-world datasets confirms that AReS effectively reveals model biases and discriminatory patterns not visible through individual recourses.
• User studies demonstrate that decision makers gain a comprehensive, actionable overview of model behavior, facilitating trust and accountability.
• Several existing individual-level recourse methods are formally shown to be special cases of the proposed framework, establishing theoretical unification.