Analytic Learning Methods for Pattern RecognitionКНИГИ » ПРОГРАММИНГ
Название: Analytic Learning Methods for Pattern Recognition Автор: Kar-Ann Toh, Huiping Zhuang, Simon Liu, Zhiping Lin Издательство: Springer Год: 2025 Страниц: 400 Язык: английский Формат: pdf (true), epub Размер: 57.9 MB
This textbook is a consolidation of learning methods which comes in an analytic form. The covered learning methods include classical and advanced solutions to problems of regression, minimum classification error, maximum receiver operating characteristics, bridge regression, ensemble learning and network learning. Both the primal and dual solution forms are discussed for over-and under-determined systems. Such coverage provides an important perspective for handling systems with overwhelming samples or systems with overwhelming parameters. For goal driven classification, the solutions to minimum classification-error, maximum receiver operating characteristics, bridge regression, and ensemble learning represent recent advancements in the literature. In this book, the exercises offer instructors and students practical experience with real-world applications.
Presently, courses in the field of Machine Learning and Artificial Intelligence (AI) place significant emphasis on Deep Learning. Such emphasis assumes that students already possess a solid foundation in the classical aspects of Machine Learning. However, essential and potent topics like analytic learning and ensemble learning seemingly receive less attention in most curricula. In particular, analytic learning offers distinct advantages, including determinism, analytical rigor, and interpretability. Unfortunately, these aspects are often overlooked. As Deep Learning systems grow increasingly complex, understanding their behavior becomes more challenging. The trend toward intricate models inevitably leads to solutions that defy human intuition and control. Given these considerations, we propose the creation of a new book specifically focused on analytic learning methods to provide readers with a comprehensive perspective in this field.
Zooming in on the analytic aspects of Machine Learning, the information appears to be relatively fragmented for the application community. For example, when it comes to linear regression, practitioners are often led to a less than satisfactory approach according to the fragmented number of references they have at hand. The over-determined system appears to be the default starting point for solving such a baseline system. When facing systems of large dimensions but with a small number of samples, such a default solution is deemed to encounter the singularity problem. The emergence of “small sample size problem” thence addressing the problem through regularization or data augmentation. Many seem to be unaware of the bigger picture of the equal-determined, over-determined, and under-determined systems to cater for the various scenarios of data with different combinations of dimension size and sample size. If such knowledge of the complete systems of linear equations was known at the beginning, the journey of solving the singularity problem would be straightforward. The first intent of this book is to fill this gap for engineering practitioners.
Moving a step forward, polynomial regression is a simple yet effective method for pattern recognition that is comparable with many nonlinear models and shallow networks. However, due to its explosive number of expansion terms for high-dimensional features, it is relatively less applied except for lower order models. By exploiting the much lower resolution of the classification target compared with that of the regression target, polynomial expansion features can be reduced while preserving their nonlinear mapping capabilities for pattern classification. This book includes a collection of polynomial expansion-based algorithms to outstretch the linear regression model for pattern classification.
- Outlines the provision of a unified treatment for data with overwhelming samples or parameters for stable predictions - Provides advanced solutions to regression and classification problems that are crucial for complex systems - Includes examples coded in Python and Matlab which provide students and instructors with mathematical insights
Скачать Analytic Learning Methods for Pattern Recognition
Навигация
Вход на сайт
Реклама
Analytic Learning Methods for Pattern Recognition 
