Lizárraga et al 2010a - Revision history

Scipediacontent at 09:56, 14 June 2017

2017-06-14T09:56:27Z

Scipediacontent: Scipediacontent moved page Draft Content 876023164 to Lizárraga et al 2010a

2017-06-14T08:43:25Z

Scipediacontent moved page Draft Content 876023164 to Lizárraga et al 2010a

Scipediacontent: Created page with "== Abstract == Una propiedad muy importante de los con- juntos no dominados es su diversidad. Mientras mayor sea la diversidad, más rica es la información sobre las posible..."

2017-06-14T07:47:52Z

Created page with "== Abstract == Una propiedad muy importante de los con- juntos no dominados es su diversidad. Mientras mayor sea la diversidad, más rica es la información sobre las posible..."

New page

== Abstract ==

Una propiedad muy importante de los con- juntos no dominados es su diversidad. Mientras mayor sea la diversidad, más rica es la información sobre las posibles soluciones a un problema multiobjetivo. Desde el inicio de la computación evolutiva multiobjetivo se han encontrado dificultades para evaluar la diversidad de los conjuntos no dominados. Muchas métricas diseñadas con este fin, fallan en ejemplos muy sencillos. En este trabajo revisamos en que consisten las principales fallas de las métricas de diversidad y damos una propuesta que no presenta estos problemas. Nuestra propuesta mide la diversidad de una forma diferente, considerando un hiper-volumen de influencia del conjunto, y tiene un comportamiento excelente como medida de desempeño. Se probró nuestra métrica usando un benchmark publicado en la bibliografía, teniendo un desempeño perfecto. Summary Diversity is a very important property for non-dominated sets. The diversity is a measure of how much information is contained in a non-dominated set. Evaluating diversity has been a diffcult issue in multi-objective evolutionary computation. Many diversity performance measures fail in simple cases. In this work, we describe the most common problems in diversity performance measures and we propose a more robust approach. The problem with most performance measures is that they consist on evaluating the standard deviation of the distances between the elements of the non-dominated sets, or a similar calculation. This dependence on a standard deviation produces a high sensibility to small changes in the non-dominated sets. Our approach is based on an hype-volume associated to the non-dominated set. The behavior of this hyper-volume is exactly what we expect from a diversity performance measure. We tested our approach using a benchmark published in bibliography, showing an exceptional performance.

== Full document ==
<pdf>Media:draft_Content_876023164RR263H.pdf</pdf>

@@ Line 1: / Line 1: @@
 == Abstract ==
-Una propiedad muy importante de los con- juntos no dominados es su diversidad. Mientras mayor sea la diversidad, más rica es la información sobre las posibles soluciones a un problema multiobjetivo. Desde el inicio de la computación evolutiva multiobjetivo se han encontrado dificultades para evaluar la diversidad de los conjuntos no dominados. Muchas métricas diseñadas con este fin, fallan en ejemplos muy sencillos. En este trabajo revisamos en que consisten las principales fallas de las métricas de diversidad y damos una propuesta que no presenta estos problemas. Nuestra propuesta mide la diversidad de una forma diferente, considerando un hiper-volumen de influencia del conjunto, y tiene un comportamiento excelente como medida de desempeño. Se probró nuestra métrica usando un benchmark publicado en la bibliografía, teniendo un desempeño perfecto. Summary Diversity is a very important property for non-dominated sets. The diversity is a measure of how much information is contained in a non-dominated set. Evaluating diversity has been a diffcult issue in multi-objective evolutionary computation. Many diversity performance measures fail in simple cases. In this work, we describe the most common problems in diversity performance measures and we propose a more robust approach. The problem with most performance measures is that they consist on evaluating the standard deviation of the distances between the elements of the non-dominated sets, or a similar calculation. This dependence on a standard deviation produces a high sensibility to small changes in the non-dominated sets. Our approach is based on an hype-volume associated to the non-dominated set. The behavior of this hyper-volume is exactly what we expect from a diversity performance measure. We tested our approach using a benchmark published in bibliography, showing an exceptional performance.
+Diversity is a very important property for non-dominated sets. The diversity is a measure of how much information is contained in a non-dominated set. Evaluating diversity has been a diffcult issue in multi-objective evolutionary computation. Many diversity performance measures fail in simple cases. In this work, we describe the most common problems in diversity performance measures and we propose a more robust approach. The problem with most performance measures is that they consist on evaluating the standard deviation of the distances between the elements of the non-dominated sets, or a similar calculation. This dependence on a standard deviation produces a high sensibility to small changes in the non-dominated sets. Our approach is based on an hype-volume associated to the non-dominated set. The behavior of this hyper-volume is exactly what we expect from a diversity performance measure. We tested our approach using a benchmark published in bibliography, showing an exceptional performance.
 == Full document ==
 <pdf>Media:draft_Content_876023164RR263H.pdf</pdf>

← Older revision	Revision as of 08:43, 14 June 2017
(No difference)