From Clustering to Cluster Explanations via Neural Networks

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Explainable AI (XAI) primarily focuses on supervised learning, leaving unsupervised clustering methods lacking interpretability.
• Existing clustering techniques identify data structures but do not explain the reasoning behind cluster assignments.
• There is a need for methods that provide feature-based explanations for unsupervised data partitioning.

Purpose of the Study:

• To develop a novel framework for explaining cluster assignments in unsupervised learning.
• To enable efficient and reliable attribution of cluster assignments to input features.
• To enhance the understanding and assessment of clustering model performance and discovered structures.

Main Methods:

• Proposed a new framework based on 'neuralizing' clustering models, transforming them into neural networks.
• Leveraged the neuralized models to efficiently and accurately attribute cluster assignments to input features.
• Demonstrated the method's capability through several showcases analyzing data and representations.

Main Results:

• The proposed framework successfully explains cluster assignments in terms of input features.
• Neuralizing clustering models allows for quick and accurate feature attribution.
• Showcases confirm the method's utility in assessing cluster quality and extracting novel insights.

Conclusions:

• This work bridges the explainability gap in unsupervised learning, specifically for clustering.
• The 'neuralization' approach offers an efficient and reliable method for understanding cluster assignments.
• The framework facilitates deeper insights into data structure and learned representations.