A kernel-based embedding framework for high-dimensional data analysis

The world is essentially multidimensional, e.g., neurons, computer networks, Internet traffic, and financial markets. The challenge is to discover and extract information that lies hidden in these high-dimensional datasets to support classification, regression, clustering, and visualization tasks. As a result, dimensionality reduction aims to provide a faithful representation of data in a low-dimensional space. This removes noise and redundant features, which is useful to understand and visualize the structure of complex datasets. The focus of this work is the analysis of high-dimensional data to support regression tasks and exploratory data analysis in real-world scenarios. Firstly, we propose an online framework to predict longterm future behavior of time-series. Secondly, we propose a new dimensionality reduction method to preserve the significant structure of high-dimensional data in a low-dimensional space. Lastly, we propose an sparsification strategy based on dimensionality reduction to avoid overfitting and reduce computational complexity in online applications

• Autores: García Vega, Sergio
• Tipo de recurso: Doctoral thesis
• Fecha de publicación: 2019
• Institución: Universidad Nacional de Colombia
• Repositorio: Universidad Nacional de Colombia
• Idioma: spa
• OAI Identifier: oai:repositorio.unal.edu.co:unal/76729
• Acceso en línea: https://repositorio.unal.edu.co/handle/unal/76729; http://bdigital.unal.edu.co/73452/
• Palabra clave: Dimensionality reduction; High-dimensional data; Kernel adaptive filtering; Embedding; Gradient descent; Online sequential learning; Sparsification; Reducción de dimensionalidad; Datos de alta dimensión; Filtrado adaptativo kernel; Incrustación; Gradiente descendente; Aprendizaje secuencial en línea; Esparsificación
• Rights: openAccess
• License: Atribución-NoComercial 4.0 Internacional