[Paper Review] A neural network and iterative optimization hybrid for Dempster-Shafer clustering

Johan Schubert|ArXiv.org|May 16, 2003

Neural Networks and Applications5 references3 citations

TL;DR

This paper proposes a hybrid clustering method that combines a neural network for fast initial clustering with iterative optimization to refine results, leveraging Dempster-Shafer theory to minimize metaconflict. The approach achieves high computational efficiency while significantly improving clustering accuracy over the neural network alone, outperforming both standalone methods in performance and speed.

ABSTRACT

In this paper we extend an earlier result within Dempster-Shafer theory ["Fast Dempster-Shafer Clustering Using a Neural Network Structure," in Proc. Seventh Int. Conf. Information Processing and Management of Uncertainty in Knowledge-Based Systems (IPMU 98)] where a large number of pieces of evidence are clustered into subsets by a neural network structure. The clustering is done by minimizing a metaconflict function. Previously we developed a method based on iterative optimization. While the neural method had a much lower computation time than iterative optimization its average clustering performance was not as good. Here, we develop a hybrid of the two methods. We let the neural structure do the initial clustering in order to achieve a high computational performance. Its solution is fed as the initial state to the iterative optimization in order to improve the clustering performance.

Motivation & Objective

• To address the trade-off between computational speed and clustering accuracy in Dempster-Shafer theory-based clustering.
• To improve upon the neural network method’s lower average performance despite its fast computation.
• To enhance the iterative optimization method’s high accuracy by reducing its computational cost through neural initialization.
• To develop a hybrid framework that leverages the strengths of both neural networks and iterative optimization for better overall clustering performance.

Proposed method

• A neural network structure is used to perform initial clustering by minimizing a metaconflict function, providing fast, approximate solutions.
• The neural network’s output is used as the initial state for iterative optimization to refine clustering results.
• Iterative optimization further minimizes the metaconflict function starting from the neural network’s solution to improve accuracy.
• The hybrid approach combines the speed of neural networks with the precision of iterative optimization in a two-stage process.
• The method operates within the framework of Dempster-Shafer theory, using belief structures to represent uncertainty in clustering.
• The metaconflict function is minimized to reduce conflict between pieces of evidence, improving clustering quality.

Experimental results

Research questions

• RQ1 Can a neural network provide a fast yet effective initial clustering solution for Dempster-Shafer theory?
• RQ2 Can iterative optimization significantly improve the clustering accuracy of a neural network’s output?
• RQ3 Does combining neural networks and iterative optimization yield better performance than either method alone?
• RQ4 How does the hybrid approach balance computational efficiency and clustering accuracy in uncertainty-based clustering?

Key findings

• The hybrid method achieves significantly better clustering performance than the neural network alone, with improved accuracy in minimizing metaconflict.
• The neural network provides a fast initial solution, reducing the number of iterations needed by the iterative optimization phase.
• The iterative optimization stage successfully refines the neural network’s output, leading to higher-quality clustering results.
• The hybrid approach outperforms both standalone methods in terms of clustering accuracy while maintaining high computational efficiency.