Рақамли технологияларнинг назарий ва амалий масалалари Volume 8 Issue 3 (2025) · pp. 101-109
Feature Extraction Methods Based on Traditional Statistics and Deep Learning
Юсупов, О.Р., Хандамов, Й.Х., Хожиакбаров, Ш.М.
Abstract
The article presents a theoretical and practical comparative analysis of four important methods related to feature extraction and dimensionality reduction: Fisher Discriminant Analysis (FDA), Canonical Correlation Analysis (CCA), Deep Canonical Correlation Analysis (DCCA), and Deep Multiset Canonical Correlation Analysis (DMCCA). The theoretical foundations, advantages, and disadvantages of each method are discussed in detail. In addition, their applicability across various domains is highlighted. The analyses show that FDA is effective in supervised learning, while CCA serves as an important tool for studying relationships between two sources. DCCA extends classical approaches through deep learning, enabling the identification of nonlinear structures. DMCCA, in turn, constructs a common latent representation for multi-source data and demonstrates high efficiency in modern artificial intelligence systems.
извлечение признаковсокращение размерностиFDACCADCCADMCCAлатентное представлениестатистические методыглубокое обучениемультимодальные данные
Metadata source: the journal's OAI-PMH archive · Sindex does not store the full text; it links to the source.