#TITLE_ALTERNATIVE#

• Main Author: (NIM 235 05 021), Nurwijaya
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/8872
• Institution: Institut Teknologi Bandung

Abstract: <br /> <br /> <br /> <br /> <br /> Many attempts have been made to imitate the way nature solves complex problems, to match its elegance, to equal its imperfect and imprecision yet ideal solutions. Two of many techniques built for the ambitious goal are Genetic Algorithms (GA) and Artificial Neural Networks (ANN). GA is reminiscent of natural selection, an elegant generate-and-test strategy known to be very effective at finding optimal or near optimal solutions to a wide variety of problem. ANN is a representation of human brain, popular for its capability to recognize patterns from noisy, complex data, and estimating their nonlinear relationships. <br /> <br /> <br /> <br /> <br /> ANN is often very difficult to design, simply because the method of forming such complex and massive network as the human brain it represents is still a great mystery. Even more, when complex combinations of performance criteria (such as learning speed, compactness, generalization ability, and noise-resistance) are given, and as network applications continue to grow in size and complexity, the traditional engineering approach will not work, a more efficient and automated solution will be needed. GA can be used to automate ANN architecture design in several ways, for example: topology optimization, genetic training algorithms, and control parameter optimization. <br /> <br /> <br /> <br /> <br /> This paper is an in-depth study of ANN optimization method by using GA, or in other words, Genetically Evolved Artificial Neural Network. The study is to be made on the benchmark result of GA optimized ANN on several tasks. <br />