Generalizability vs. Counterfactual Explainability Trade-Off

2025-05-29

AI Summary

This paper identifies a fundamental trade-off between generalization performance and counterfactual explainability in supervised learning: poorer generalization (i.e., stronger overfitting) facilitates easier generation of counterfactual examples. Method: We propose a novel metric—ε-effective counterfactual probability (ε-VCP)—grounded in the geometric properties of decision boundaries and local input perturbations. We theoretically prove that ε-VCP monotonically increases with overfitting severity, establishing it as a reliable proxy for generalization degradation. Contribution/Results: This work establishes, for the first time, a formal theoretical connection between generalization and counterfactual explainability. Empirical validation across multiple benchmark datasets confirms that ε-VCP consistently tracks declining generalization performance. As such, ε-VCP serves as a quantifiable, interpretable diagnostic tool for both model robustness assessment and explainability evaluation.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Reasoning under Uncertainty: Causality

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

In this work, we investigate the relationship between model generalization and counterfactual explainability in supervised learning. We introduce the notion of $varepsilon$-valid counterfactual probability ($varepsilon$-VCP) -- the probability of finding perturbations of a data point within its $varepsilon$-neighborhood that result in a label change. We provide a theoretical analysis of $varepsilon$-VCP in relation to the geometry of the model's decision boundary, showing that $varepsilon$-VCP tends to increase with model overfitting. Our findings establish a rigorous connection between poor generalization and the ease of counterfactual generation, revealing an inherent trade-off between generalization and counterfactual explainability. Empirical results validate our theory, suggesting $varepsilon$-VCP as a practical proxy for quantitatively characterizing overfitting.

Problem

• Investigates trade-off between model generalization and counterfactual explainability
• Introduces ε-VCP to measure counterfactual probability near decision boundaries
• Links poor generalization to easier counterfactual generation via ε-VCP

Innovation

• Introduces $varepsilon$-VCP for explainability analysis
• Links decision boundary geometry to overfitting
• Proposes trade-off between generalization and explainability