Immune Optimization Algorithms And Applications Based On Cluster Computing

Immune optimization algorithms are one of important research aspects of artificial immune system, which has been widely used in many fields. As a new intelligent search algorithm, there still exist some open problems in the theories and the applications, such as solution quality and speed. With the rapid development of parallel software and hardware platforms, parallel algorithms are capable of providing a new solution for the problems. Many new algorithms and strategies are proposed for feature selection and bin packing based on different parallel platforms, and the main work can be summarized as follows:(1)Focusing on the problem of dimensionality reduction in data mining and pattern recognition, a novel algorithm for feature selection is proposed based on antibody clonal selection and immune memory principle. Antibody population is used for global exploration,and memory unit which only reserves the best individuals with embedded local search operations is designed for fine-tune search. Different fitness functions for antibody population and memory unit are used to improve search performance. The fitness of an individual is determined by evaluating the nearest neighbor classifier with leave-one-out cross-validation. Experimental results on several high-dimensional standard datasets show that the proposed algorithm outperforms conventional genetic algorithm and classical sequential floating forward search algorithm in terms of classification accuracy and robustness..(2)A novel parallel immune clonal selection algorithm is proposed for feature selection based on clonal selection theory in an artificial immune system and message passing interface. The presented method uses an immune clonal selection which can combines global exploration and fine-tune search for feature selection; Fitness of feature subset fitness is determined by evaluating the nearest neighbor classifier with leave-one-out cross-validation. In order to reduce running time, fitness of antibody is evaluated by many processors in a cluster based on a master-slave algorithm. Based on LPT and MULTIFIT algorithms, we present a heuristic algorithm for tasks scheduling in nodes. Experimental results on 3 standards UCI dataset sets show that the proposed algorithm outperforms standard genetic algorithm and classical sequential floating forward search algorithm in terms of classification accuracy; we achieved 75 percent efficiency even when up to 40 processors were used.(3)To classify high dimensional data produced from a multiresolution and multidirection image representation must deal with the curse of dimensionality. A novel parallel immune clonal feature selection algorithm for dimensionality reduction is proposed based on biological immune clone theory and message passing interface. Multiple anti-populations distributed in multi-processors are used to search for feature subset, to avoid premature convergence and preserve the population diversity; populations are linked in a ring topology and communicate by an adaptive vaccine migration operations. The features are extracted using energy information obtained from Contourlet transform. Feature subset is evaluated by the nearest neighbor classifier with leave-one-out cross-validation. Experimental results on Brodatz textures dataset and real SAR images show that the proposed algorithm outperforms conventional genetic algorithm and classical sequential floating forward search algorithm and greatly reduce running time using Linux cluster, we achieved 91 percent efficiency even when up to 13 processors were used.(4)This paper deals with the one-dimensional cutting stock problem and bin packing problem and presents a parallel metaheuristic solution approach based on the Artificial Immune System. A novel parallel immune clonal selection algorithm for bin packing problem is proposed based on biological immune clone theory and message passing interface. Multiple anti-populations distributed in multi-processors are linked in a ring topology and communicate by an adaptive migration operations. Experimental results on cutting stock datasets and OR-library show that the proposed algorithm outperforms classical algorithms and greatly reduce running time using Linux cluster.