QTM-GMM (MODIFICATION OF THE GMM ALGORITHM USING PARAMETERIZED EXPONENTS), 2024

Project leader: Sergey Koltsov

Participants: Anton Surkov, Ksenia Kupitman

The project to optimize the GMM (Gaussian Mixture Model) algorithm using parameterized functions was completed in 2024 (a continuation of the QTM project completed in 2023). As part of the project, the traditional GMM model, based on a mixture of Gaussian distributions, was expanded by introducing q-parameterized Gaussians - modified versions of distributions adapted through the q parameter. This made it possible to increase the flexibility of the model when working with heterogeneous data. Parameterized Renyi entropy was used as a criterion for determining the optimal number of clusters, which, in combination with q-Gaussians, ensured more accurate detection of hidden patterns and structures in complex sociological and psychological data.