Abián Hernández, H. Fabelo, S. Ortega, Abelardo Báez, G. Callicó, R. Sarmiento
{"title":"Random forest training stage acceleration using graphics processing units","authors":"Abián Hernández, H. Fabelo, S. Ortega, Abelardo Báez, G. Callicó, R. Sarmiento","doi":"10.1109/DCIS.2017.8311636","DOIUrl":null,"url":null,"abstract":"Graphics Processing Units (GPUs) are platforms very appropriated to accelerate processes with high computational load, like the supervised classification of hyperspectral images. The supervised classifier Random Forest has proved to be a good candidate to classify hyperspectral images and currently constitutes an emerging technology for medical diagnosis. The objective of this paper is focused in the Random Forest training phase acceleration using GPUs, starting from an efficient CPU implementation. For some applications, it is necessary to refine the classification model depending on the new acquired samples. In this paper are presented solutions for two bottlenecks identified in the training stage in order to accelerate the algorithm. The different solutions for the bottlenecks provided in this research study have demonstrated that GPU implementation is a promising technique to generate models in shorter time. With this implementation it is possible to achieve the training process in real-time.","PeriodicalId":136788,"journal":{"name":"2017 32nd Conference on Design of Circuits and Integrated Systems (DCIS)","volume":"29 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"5","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2017 32nd Conference on Design of Circuits and Integrated Systems (DCIS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/DCIS.2017.8311636","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 5
Abstract
Graphics Processing Units (GPUs) are platforms very appropriated to accelerate processes with high computational load, like the supervised classification of hyperspectral images. The supervised classifier Random Forest has proved to be a good candidate to classify hyperspectral images and currently constitutes an emerging technology for medical diagnosis. The objective of this paper is focused in the Random Forest training phase acceleration using GPUs, starting from an efficient CPU implementation. For some applications, it is necessary to refine the classification model depending on the new acquired samples. In this paper are presented solutions for two bottlenecks identified in the training stage in order to accelerate the algorithm. The different solutions for the bottlenecks provided in this research study have demonstrated that GPU implementation is a promising technique to generate models in shorter time. With this implementation it is possible to achieve the training process in real-time.