DiCoFlex: Model-agnostic diverse counterfactuals with flexible control

2025-05-29

AI Summary

Existing counterfactual explanation methods rely on black-box model access, require per-sample optimization, and lack flexibility in accommodating user-defined constraints (e.g., sparsity, actionability). This paper proposes the first model-agnostic conditional normalizing flow framework that generates multiple diverse counterfactuals satisfying heterogeneous constraints via a single forward pass—eliminating gradient-based optimization and model retraining entirely. Our approach integrates supervised flow modeling with constraint-aware sampling, enabling real-time, optimization-free customization of constraints. Evaluated on standard benchmarks, our method significantly improves counterfactual validity, diversity, proximity, and constraint satisfaction rates. It delivers an efficient, controllable, and deployable explainable AI solution for high-stakes decision-making scenarios.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories
• Computer Vision: Interpretability, Explainability, and Transparency

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
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Counterfactual explanations play a pivotal role in explainable artificial intelligence (XAI) by offering intuitive, human-understandable alternatives that elucidate machine learning model decisions. Despite their significance, existing methods for generating counterfactuals often require constant access to the predictive model, involve computationally intensive optimization for each instance and lack the flexibility to adapt to new user-defined constraints without retraining. In this paper, we propose DiCoFlex, a novel model-agnostic, conditional generative framework that produces multiple diverse counterfactuals in a single forward pass. Leveraging conditional normalizing flows trained solely on labeled data, DiCoFlex addresses key limitations by enabling real-time user-driven customization of constraints such as sparsity and actionability at inference time. Extensive experiments on standard benchmark datasets show that DiCoFlex outperforms existing methods in terms of validity, diversity, proximity, and constraint adherence, making it a practical and scalable solution for counterfactual generation in sensitive decision-making domains.

Problem

• Generates diverse counterfactuals without constant model access
• Enables real-time user-driven constraint customization
• Improves validity, diversity, and adherence to constraints

Innovation

• Model-agnostic diverse counterfactuals generation
• Conditional normalizing flows for flexibility
• Real-time user-driven constraint customization