Scalable Propabilistic Approximation method in Applications

Abstract

This thesis aims to extend the research on the newly developed Scalable Probabilistic Approximation (SPA) method, with emphasis predominantly on classification problems. The SPA method is utilized to discretize continuous stochastic processes and, in conjunction with Bayesian causal inference modeling, leads to a multiobjective optimization problem that is capable of simultaneously resolving both objectives. The solution to this problem is formulated as a supervised machine learning algorithm that is suitable for various classification tasks. Although the algorithm is limited in terms of computational cost, a proposed estimation of the problem, which is closely related to the widely known K-means algorithm, is applicable even for large datasets. Preliminary experiments demonstrate that this framework is adaptable to the selected application of corrosion detection from image data.