|
PRI06A227 - Optimization of classification systems using GA and BOINC architecture |
|
| Title: | Optimization of classification systems using genetic algorithms and BOINC architecture |
|---|
Financing agency:
| Junta de Extremadura
|
|---|
From/to: | January 2007 - December 2008 |
|---|
Coordinator: | Carlos J. García Orellana
|
|---|
| CAPI members: | 5
|
|---|
| Description: | The objective of this project is to evaluate the usage of the telematic and computing infrastructure present in the Institutions of Extremadura (schools, hospitals, university, administration) in order to solve scientific problems with a high computational cost. Our proposal consists in studying the viability and outlining solutions to the problems of this kind of computing in complex optimization processes.
To do that, we consider the development of SoftComputing tools, to be used in PRC (Public Resource Computing) environments. The project is based in the usage of BOINC (Berkeley Open Infrastructure for Network Computing) architecture, because this will be the architecture present in Extremadura.
Particularly, we propose the adaptation and development of tools in order to approach the optimization of classification systems by means of Genetic Algorithms, usually in problems of image analisys. The optimization of classification systems is a complex problem, with a large number of parameters to adjust (in the characteristics extraction and selection, and in the classifier itself) and normally with an enormous computational cost and low communication load. This high ratio computation/communication makes the problem suitable for the BOINC architecture.
To carry out the project, we propose three sequential objectives: first we will check the operation of neuronal classifier simulations under BOINC (we already have the application migrated); second we will approach the execution of Genetic Algorithms (GA) to optimize problems with high computational cost. As a third objective we consider the inclusion of a model based in the concept of “Quality of Service” (QoS), into the simulation process, using dedicated and controlled resources, i.e., using an small helping cluster along with BOINC.
The practical applications of this project are numerous. During the period of the project we will focus in the application to a couple of problems with which our research group is working: classification of cloud cover and classification of microcalcification clusters in mammograms. If the project yields satisfactory results, the developed system will be very useful for approaching other complex xlassification problems. But that would be only the beginning, the tools will be designed and developed having in mind its application to other optimization problems, and will be made publicly availables. This implies that other research groups will be able to use, and benefit from, the developed tools in their own problems, taking advantage of the experience acquired by our group.
The fact that this project can be the base for others, is an important motivation for its proposal, because to maintain such an infrastructure only makes sense if there are many users for it. Because of this, we intend to provide the infrastructure that the Junta de Extremadura is introducing with initial contents and applications, and to show to others the potential of this novel BOINC-based technology.
|
|---|
|
Research