high performance computing on graphics processing units: hgpu.org

This paper introduces a Multi Objective Parallel Genetic Algorithm (MOPGA) using the Compute Unified Device Architecture (CUDA) hardware for parallel processing. The algorithm demonstrates significant speed gains using affordable, scalable and commercially available hardware. The algorithm implements a document search using techniques such as Term Frequency Inverse Document Frequency (TF-IDF), Latent Semantic Analysis (LSA), Multi Objective Algorithms (MOA), Genetic Algorithm (GA), and Quad Tree Pareto Dominance techniques. The objective of the proposed algorithm is to assemble an adaptable and scalable search mechanism to efficiently retrieve highly relevant document for a given search query. TFIDF and LSA vector space searches are two of the more common approaches to text mining. We have demonstrated that by combining results from both operations the number and quality of results could be improved. Evolutionary algorithms, specifically Genetic Algorithms have long been used to efficiently optimize multi-objective problems and so provide a natural starting point for our approach.