Energy-efficient Parallel and Distributed Multi-objective Feature Selection on Heterogeneous Architectures
Year: 2020
Type: dissertation
Abstract: The objectives and requirements of computer science are constantly \nchanging. Nowadays, when developing an algorithm, it is not enough to \nsolve the problem itself since the energy-time performance or memory \nusage should also be taken into account, especially in high-dimensional \nproblems such as FS. For some years, energy-aware computing has \ngained importance as it allows data centers to save costs by reducing \nenergy consumption, and it is even today a topic of global interest due \nto environmental reasons. The present trend in the development of \ncomputer architectures that offer improvements in both performance \nand energy efficiency has provided distributed platforms with interconnected \nnodes including multiple multi-core CPUs and accelerators. \nIn these so-called heterogeneous systems, the applications can take \nadvantage of different parallelism levels according to the characteristics \nof the devices in the platform. Precisely, these differences between \ncomputing devices are what make heterogeneous computing, unlike \nhomogeneous, present other problems to deal with. However, this process \nis not automatic and requires the intervention of the developer to \nproperly program the applications and thus achieve good results. \nWith this in mind, the objective of this thesis is the development of \nparallel and energy-efficient codes for time-demanding problems that \nfrequently appear in bioinformatics and biomedical engineering applications. \nSpecifically, this thesis tackles with unsupervised EEG classification, \nwhich is one of the aforementioned high-dimensional problems \ndue to the characteristics of the EEG signals. To cope with the high \nnumber of features that each EEG contains, the implemented procedures \nare based on a multi-objective FS approach. The codes have \nbeen designed to take advantage of the heterogeneous architectures by \nexploiting the computing capabilities of their devices. In addition, they \nhave been developed in a procedural way due to their complexity. This \nthesis also studies and compares the codes to identify the advantages \nand drawbacks of each, as well as analyzes the performance behavior in \nterms of energy consumption, execution times, and quality of the solutions \nunder different situations such as workload, available computing \nresources, device clock frequency, and others that will be described \nin the corresponding chapters. The results show the importance of \ndeveloping efficient methods to meet the energy-time requirements, \npointing out the methodology to be followed and demonstrating that \nenergy-aware computing is the way to continue on the right track.
Authors: Escobar Pérez, Juan José
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3115199346