ExplainabilityPrivacyTransparencycs.LG
Contrastive Explainable Clustering with Differential Privacy
Dung Nguyen, Ariel Vetzler, Sarit Kraus, Anil Vullikanti
June 7, 2024
Adaptive Agents and Multi-Agent Systems
4 citations
Key Takeaways
-
Privacy and explainability can coexist without major trade-offs: The paper demonstrates that adding differential privacy protections to clustering explanations maintains essentially the same utility bounds as non-private explanations, suggesting organizations don't need to choose between protecting individual data and providing understandable AI decisions.
-
Contrastive explanations offer personalized insights: Rather than generic explanations, this approach calculates how clustering decisions change when specific data points are involved, providing individuals with tailored understanding of why they were grouped together—relevant for applications like customer segmentation or resource allocation.
-
The method works across real-world datasets: Experiments show the approach produces "meaningful, privacy-preserving, and individually relevant explanations" across various datasets, indicating practical applicability beyond theoretical scenarios.
-
Addresses a gap in responsible AI: By combining three important elements—explainability, privacy protection, and personalization—this work tackles a real challenge in deploying AI systems that are both transparent to users and protective of their data.
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.