Contrastive explainable clustering with differential privacy

2024-06-07

arXiv.org

AI Summary

Balancing interpretability and differential privacy in clustering remains challenging, particularly for k-means/k-median algorithms. Method: We propose the first personalized contrastive explanation framework tailored to k-means/k-median, anchoring explanations at the target data point as a pseudo-centroid and quantifying its influence on clustering utility. To ensure ε-differential privacy, we introduce a utility-sensitive noise injection mechanism. Contribution/Results: We theoretically prove that, under differential privacy constraints, the contrastive explanation utility bound of our framework converges to the non-private optimum—achieving, for the first time, a rigorous trade-off among privacy preservation, explanation fidelity, and individual relevance. Extensive experiments across multiple benchmark datasets demonstrate that our method incurs less than 1.5% clustering utility loss while limiting explanation error increase to within 3%, significantly outperforming existing differentially private explanation baselines.

Technology Category

• Machine Learning: Privacy
• Data Mining & Knowledge Management: Representing, Reasoning, and Using Provenance, Trust
• Computer Vision: Interpretability, Explainability, and Transparency

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Security and Privacy: Data transparency and provenance
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

This paper presents a novel approach to Explainable AI (XAI) that combines contrastive explanations with differential privacy for clustering algorithms. Focusing on k-median and k-means problems, we calculate contrastive explanations as the utility difference between original clustering and clustering with a centroid fixed to a specific data point. This method provides personalized insights into centroid placement. Our key contribution is demonstrating that these differentially private explanations achieve essentially the same utility bounds as non-private explanations. Experiments across various datasets show that our approach offers meaningful, privacy-preserving, and individually relevant explanations without significantly compromising clustering utility. This work advances privacy-aware machine learning by balancing data protection, explanation quality, and personalization in clustering tasks.

Problem

• Combines contrastive explanations with differential privacy for clustering
• Evaluates utility difference between original and fixed-centroid clustering
• Balances privacy protection, explanation quality, and personalization

Innovation

• Combines contrastive explanations with differential privacy
• Calculates utility difference for personalized centroid insights
• Achieves privacy without compromising clustering utility