Probabilistic Metric Based Multidimensional Scaling
Mika Sato‐Ilic · Procedia Computer Science · 2020
Multidimensional scaling is a well-known dimensionality reduction method and useful for summarizing the latent vital structure of the complex data. However, conventional multidimensional scaling is based on Euclidean distance between a pair of objects; there is a limitation to adjust the model to various similarity structures of real data. Therefore, this paper proposes a probabilistic metric based multidimensional scaling. For the probabilistic metric, t-norms in a probabilistic metric space and a kernel-based metric are employed. Several numerical examples are demonstrated to show a better performance of the proposed method.