Computation methods using Graphic Processing Units (GPU) for solving parallelizable problems have attracted many research interest in recent years. Following up this trend, implementations of Genetic Algorithms (GA) to GPU have been reported based on parallelism in computation tasks of population in several researches. This paper proposes an implementation method of GA on the CUDA environment, which is a general purpose computation environment for GPUs provided by NVIDIA, adopting not only parallelization of population but that of individuals. The performance of proposed implementation method are compared to a CPU implementation by the computation time using test functions and an Evolutionary Robotics problem. The proposed implementation method generated 7.6-23.0 times faster results than those of a CPU implementation.
Recently, a lot of evolutionary computation methods such as GAs for multiobjective optimization problems have been reported, and then many objective optimization problems, which have a lot of objective functions, have been especially focused on with the background of the improvement of computer performance. One of the goals in multiobjective optimization problems is to obtain various solutions superior to other solutions in at least one objective function. However, the solutions which are superior in an objective function but not in others are not what users need in practical problems, which makes the multipoint search of MOGA ineffective. This paper investigates the method which enables a user to find practical solutions for him/her by changing the direction of the search interactively, and studies the effectiveness of the proposed method through Nurse Scheduling Problem.
We propose an Interactive Genetic Algorithm(IGA) applying paired preference test that obtains evaluations of individuals from multiple people. Many Interactive Evolutionary Computation(IEC) systems evaluate the solution using only a single person. Therefore, solutions to which numerous people consent cannot be obtained. To solve this problem, we propose a multiple people participation system like IGA system that evaluates solutions using Kansei evaluations from numerous people's votes. In previous research, IGA with tournament evaluation by multiple people proposed as a basic model that obtains evaluations of individuals from multiple people's votes. And the effectiveness of the method was demonstrated by numerical simulation. However, when it assumed vote acquisition by real Web system, estimation of the number of the users participating in a vote is difficult. Therefore, in the conventional method, there are problems that the system is not acquired the number of the votes that is necessary for solution evaluation and cannot let tournament evaluation perform. To solve these problems, the effective use of the votes is requested. Then, we solve these problems by applying paired preference test that statistical technique to the conventional method and try promotion of the efficiency of the tournament evaluation. It is thought that the proposed system can decide the victory or defeat of the solution candidates at the early stage of the vote by applying paired preference test. The effectiveness of the proposed technique was verified by a numerical simulation using multiple evaluation agents instead real users. The simulation results shows that the vote necessary for the solution evaluation decreased about 80% in the proposed technique. Moreover, we compare the proposed technique with a conventional method that IGA with tournament evaluation by multiple people. we confirmed that the proposed technique is effective for the efficiency of the tournament competition.
1. Purpose In the various manufacturing fields including the building field, the form design is devised by the designer. However, the designer's idea range depends on his personal background such as education or environment, so it has been pointed out that his design idea is controlled by his background. The facade of the building controls the visitor's impression, so the image of the design is very important. Various studies such as the constitutions or histories of the facade are performed, but the facade design support system has not been developed. Therefore, if there is an idea generation support system which is defeating a conventional idea of the designer and expresses the detail of the facade, it would be effective. The purpose of this study is to develop the idea generation support system giving “discovery” to the designer for the facade design of an office building. As the first step, this study analyzes the form elements constituting a facade and develops the IDE system reducing the user's psychological burden. The effectiveness of IDE is verified by comparing three different cases of the number of individuals in IDE, for the user's psychological burden and satisfaction degree as the idea generation support system. 2. Methods 58 form elements are extracted by collecting and classifying the image of an existing office building. 58 form elements are replaced to the individual's vector of IDE, and became incorporated into IDE algorithm. 3D image of facade by IDE algorithm is expressed by VRML. The verification of this system was performed by 10 students learning architecture. 3. Conclusions In IDE algorithm, the case that the number of individuals is larger than the number of evaluation individuals is more effective than the case that the number of individuals is equal to the number of evaluation individuals. And, the individuals except evaluation individuals do not need to be changed on each generation. In this system, the number of evaluation individuals is not changed but the number of individuals is changed. Therefore, the evaluation of the whole system is improved, because user's psychological burden does not occur and the form variations are increased. In form elements, there are small differences by evaluator's personal background, but it seems to be proper because exciting designs are generated for about 90% of all the trials, and the end condition is realized in about half of the trials. The number of data of verification is small, but this system seems to be effective because this system can offer an exciting design to all evaluators in a comparatively early generation.
A wireless sensor network, which is a key network to facilitate ubiquitous environments, has attracted a significant amount of interest from many researchers. In a wireless sensor network, flooding is required for the dissemination of queries and event announcements. The original flooding causes the overlap problems. In the original flooding, generally, all sensor nodes receiving a broadcast message forward it to its neighbors by the full forwarding power. For a dense wireless sensor network, the impact caused by the original flooding may be overwhelming. The original flooding may result in the reduced network lifetime. To obtain plural adjustment solutions on the forwarding power of each sensor node, this paper proposes a new query dissemination method based on the advanced particle swarm optimization algorithm computing plural acceptable solutions. By using the obtained plural adjustment solutions, the flexible operation according to the residual energy of each sensor node can be realized. The proposed method is evaluated by numerical experiments. In the experiments performed, the performance of the proposed method is compared with those of the existing ones to verify its effectiveness.