Deploying massive runs of evolutionary algorithms with ECJ and Hadoop: Reducing interest points required for face recognition. (September 2018)

• Record Type: Journal Article
• Authors: Chávez, Francisco; Fernández de Vega, Francisco; Lanza, Daniel; Benavides, César; Villegas, Juan; Trujillo, Leonardo; Olague, Gustavo; Román, Graciela
• Other Names: Bland Wesley guest-editor.; Erez Mattan guest-editor.; Hidalgo J Ignacio guest-editor.; Fernández de Vega Francisco guest-editor.; Mercier Guillaume guest-editor.
• Is Part Of: International journal of high performance computing applications. Volume 32:Number 5(2018)
• Page Start: 706
• Page End: 720
• Publication Date: 2018-09
• Subjects: ECJ -- face recognition -- Hadoop -- parallel evolutionary algorithm; High performance computing -- Periodicals; Supercomputers -- Periodicals; 004.1105
• DOI: 10.1177/1094342016678302
• Languages: English
• ISSNs: 1094-3420

Abstract

In this paper we present a new strategy for deploying massive runs of evolutionary algorithms with the well-known Evolutionary Computation Library (ECJ) tool, which we combine with the MapReduce model so as to allow the deployment of computing intensive runs of evolutionary algorithms on big data infrastructures. Moreover, by addressing a hard real life problem, we show how the new strategy allows us to address problems that cannot be solved with more traditional approaches. Thus, this paper shows that by using the Hadoop framework ECJ users can, by means of a new parameter, choose where the run will be launched, whether in a Hadoop based infrastructure or in a desktop computer. Moreover, together with the performed tests we address the well-known face recognition problem with a new purpose: to allow a genetic algorithm to decide which are the more relevant interest points within the human face. Massive runs have allowed us to reduce the set from about 60 to just 20 points. In this way, recognition tasks based on the solution provided by the genetic algorithm will work significantly quicker in the future, given that just 20 points will be required. Therefore, two goals have been achieved: (a) to allow ECJ users to launch massive runs of evolutionary algorithms on big data infrastructures and also (b) to demonstrate the capabilities of the tool to successfully improve results regarding the problem of face recognition.