GPU-based biclustering for neural information processing

Abstract

This paper presents an efficient mapping of geometric biclustering (GBC) algorithm for neural information processing on Graphical Processing Unit (GPU). The proposed designs consist of five different versions which extensively study the use of memory components on the GPU board for mapping the GBC algorithm. GBC algorithm is used to find any maximal biclusters, which are common patterns in each column in the neural processing and gene microarray data. A microarray commonly involves a huge number of data, such as thousands of rows by thousands of columns so that finding the maximal biclusters involves intensive computation. The advantage of GPU is its ability of parallel computing which means that for those independent procedures, they can be carried out at the same time. Experimental results show that the GPU-based GBC could reduce the processing time largely due to the parallel computing of GPU, and its scalability. As an example, GBC algorithm involves a large number of AND operations which utilize the parallel GPU computations, that can be further practically used for other neural processing algorithms. © 2012 Springer-Verlag.