T. zhou, J. M. Mira, J. Cara
Self-Organizing Maps (SOM) are widely used for dimensionality reduction and clustering while preserving topological relationships of high-dimensional data in a lower-dimensional space. However, the reliability of SOM is influenced by factors such as initialization sensitivity, convergence stability, and noise in data representation. This study investigates the uncertainty in SOM training and its impact on topology preservation, analyzing how variations in initialization, learning parameters, and data perturbations affect the final mapping quality. Through empirical evaluation and theoretical exploration, we assess the robustness of SOM in maintaining neighborhood structures under different conditions. The findings contribute to a deeper understanding of SOM's reliability in practical applications.
Keywords: SOM, UNCERTAINTY
Scheduled
Classification and pattern recognition
June 11, 2025 3:30 PM
Auditorio 1. Ricard Vinyes