TDCC: A Trustworthy Deep Credal Clustering Method for Uncertain Data.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Deep clustering models often exhibit overconfidence, misclassifying ambiguous data points.
• Existing methods struggle to quantify and manage uncertainty in cluster assignments.

Purpose of the Study:

• To develop a trustworthy deep credal clustering framework that quantifies uncertainty.
• To improve the robustness and accuracy of deep clustering models.

Main Methods:

• Integration of deep neural networks with the Dempster-Shafer Theory of evidence (DST).
• Leveraging credal cluster structures to handle uncertain data.
• Derivation of closed-form solutions for cluster membership and prototype updates using coordinate descent optimization.

Main Results:

• The proposed method effectively refrains from assigning uncertain samples, reducing errors.
• Demonstrated enhanced overall clustering effectiveness across various datasets.
• Improved model trustworthiness by managing ambiguity in data.

Conclusions:

• Trustworthy deep credal clustering offers a robust solution for handling uncertainty in deep learning models.
• The framework enhances data clustering by acknowledging and managing sample ambiguity.
• This approach leads to more reliable and accurate clustering outcomes.