Hongbing Zhu

Euclidean Particle Swarm Optimization

Particle swarm optimization (PSO) is a swarm intelligence algorithm, has been successfully applied to many engineering optimization problems and shown its high search speed in these applications. However, as the dimension and the number of local optima of optimization problems increase, PSO and most existing improved PSO algorithms such as, the standard particle swarm optimization (SPSO) and the Gaussian particle swarm optimization (GPSO), are easily trapped in local optima. In this paper we proposed a novel algorithm based on SPSO called Euclidean particle swarm optimization (EPSO) which has greatly improved the ability of escaping from local optima. To confirm the effectiveness of EPSO, we have employed five benchmark functions to examine it, and compared it with SPSO and GPSO. The experiments results showed that EPSO is significantly better than SPSO and GPSO, especially obvious in higher-dimension problems.

Protein Structure Prediction with EPSO in Toy Model

Predicting the structure of protein through its sequence of amino acids is a complex and challenging problem in computational biology. Though toy model is one of the simplest and effective models, it is still extremely difficult to predict its structure as the increase of amino acids. Particle swarm optimization (PSO) is a swarm intelligence algorithm, has been successfully applied to many optimization problems and shown its high search speed in these applications. However, as the dimension and the number of local optima of problems increase, PSO is easily trapped in local optima. We have proposed an improved PSO algorithm is called EPSO in the other paper, which has greatly improved the ability of escaping form local optima. In this paper we applied EPSO to the structure prediction of toy model both on artificial and real protein sequences and compared with the results reported in other literatures. The experimental results demonstrated that EPSO was efficient in protein structure prediction problem in toy model.

Paralleling Clonal Selection Algorithm with OpenMP

Clonal selection algorithm (CSA) is one of the most representative Immune algorithms (IA) and was applied into the protein structure prediction (PSP) on AB off-lattice model, but it required a long time in the calculation. So in this paper, a parallel clonal selection algorithm (CSA) was proposed, which was implemented using distributed computing model that employed Open MP on four core computer. In the algorithm, several sub-populations replaced the original single population, and each sub-population evolved independently, and the current best individual was distributed into all the sub-populations. The parallel algorithm overcame premature convergence and found global optima efficiently. And the experiment results shown that the performance had beensignificantly improved.

Clonal Selection Algorithm with Aging Operators for Protein Structure Prediction on AB Off-Lattice Model

Immune algorithms (IAs) are microscopic view of evolutionary algorithms (EAs) and applied in combinatorial optimization problems. This paper addresses to a clonal selection algorithm (CSA) that is one of the most representative IA and was applied into the protein structure prediction (PSP) on AB off-lattice model, in which the memory B cells of the CSA was innovated by employing different strategies: local search and global search in the phase of the mutation. And the CSA was further improved by adding aging operator to combat the premature convergence. However the pure aging operator didn’t achieve effective results and sometimes the optimum solution was eliminated. To resolve this problem, the current best solution was reserved by an antibody and it was not eliminated when its age reached its life span. In our experiments the improved algorithm was compared with the standard CSA and the pure aging CSA, which of the results demonstrated that the improved strategy with the memory B cells and long life aging was very effective to overcome premature convergence and to avoid trapped in the local-best solution, and it was also effective in keeping the diversity of the small size population.

Structure Optimization by an Improved Tabu Search in the AB Off-Lattice Protein Model

Tabu search is a meta-heuristic approach that is found to be useful in solving combinatorial optimization problems. This paper employs the adaptive memory features of tabu search to deal with protein folding problem. A kind of optimization of the neighborhood scale is presented, where a annealing mechanism is also used to enhance the searching ability for optimum solutions of the AB off-lattice model. This model has only two types of residues: hydrophobic (A) and hydrophilic(B). Based on the AB off-lattice model, the problem is converted from a nonlinear constraint-satisfied problem to an unconstrained optimization problem. Experimental results demonstrate that the proposed methods are very promising for searching the ground states of protein folding in two dimensions.

Implementation of 3D SRAD Algorithm on CUDA

Speckle reducing an isotropic diffusion (SRAD) for three dimensional (3D) ultrasound images is an an isotropic diffusion approach for smoothing speckled imagery and is based on the conventional an isotropic diffusion and the traditional speckle reducing filter. The 3D SRAD has improved the quality of edge preservation and the smoothness of homogenous regions, but it requires much time to complete complex computation that affects its real-time application. To resolve this problem, one parallel 3D SRAD (3D pSRAD) was proposed on CUDA that is a general purpose parallel computing architecture and has solved many complex computational problems efficiently. In additional, a lot of experiments based on real images were employed to demonstrate the effectiveness and availability for the algorithm. The experimental results revealed a 60X parallel speedup over the original 3D SRAD.

Paralleling Euclidean Particle Swarm Optimization in CUDA

Euclidean Particle Swarm Optimization (EPSO) is a swarm intelligence algorithm, which has been successful applied to many engineering optimization problems and shown its high search speed in these applications. However, with the increase in the dimension of optimization problems and the number of local optima, the processing speed of the EPSO has become a bottleneck of applications as each particle of them has to calculate separately fitness. In this paper the EPSO has been parallelled in Compute Unified Device Architecture (CUDA) to solve the bottleneck. Five benchmark functions had been employed to examine the performance of the parallelled EPSO (pEPSO), and the experimental results shown that the average processing of calculating fitness had been accelerated to maximum 16.27 times the original algorithm.

Parallelism of Clonal Selection for PSP on CUDA

Protein structure prediction (PSP) is the process of searching for the min energy of the protein. While many algorithms have being put forward to predict the structure of protein, the complicated computation make the time cost of the algorithms are significantly expensive. CUDA, the newly developing technology, makes us use Graphic Processing Unit (GPU) that is mainly used on Graphic processing at before, much better to develop the general parallel program. So, it gives us a new direction to study the parallelism on PSP. In this paper, we select the clonal selection algorithm for PSP on AB Off-Lattice model and have attempted to fulfill it on the CUDA platform. The results from the experiments showed that it reduced the time cost of the algorithm greatly with a suitable precision. It also provides us a new beginning of the research on PSP.

Paralleling Genetic Annealing Algorithm with OpenMP

In this paper, a parallel Genetic Annealing Algorithm (GAA) combining with simulated annealing algorithm and genetic algorithms was proposed, which was implemented using distributed computing model that employed OpenMP on four core computer. In this algorithm, several sub-populations replaced the original single population, and each sub-population evolved independently, and the current best individual was distributed into all the sub-populations. The algorithm overcame premature convergence, and found global optima efficiently in less time. And the experiment results shown that the performance had been significantly improved.

Paralleling Genetic Annealing Algorithm on Grid

Protein structure prediction (PSP) aims to search min energy for protein, which is a problem of non-deterministic polynomial (NP). A lot of algorithms have been proposed to solve this problem, and the time cost of the algorithms are significantly expensive due to complex computation. Genetic annealing algorithm (GAA) combining with simulated annealing algorithm and genetic algorithm, which is one of the most representative algorithms to be applied into the PSP, also requires large computing power for the complexity of the algorithm itself. To improve the efficiency of the algorithm, we established a grid system, implemented one parallel GAA algorithm in this system, and made some improvements to this algorithm. The results of the experiments shown that the parallel GAAs maximum speed-up rate is 3.81 times the serial algorithm.