Genetic Programming Techniques for Modelling and Design

点击翻译按钮获取中文摘要

Computational intelligence is a branch of artificial intelligence that uses problem-solving algorithms inspired by phenomena seen in nature. My research is in genetic programming (GP). This is a computational intelligence technique inspired by Darwin's theory of evolution, which is used to automatically "breed" computer programs to solve a variety of problems. I am interested in the use of GP to solve problems in evolutionary design. This is a research area involving complex design problems found in diverse areas such as 3D modelling, architecture, art, biological networks, and others. A common theme in design fields is the assumption that solutions inherently require a high degree of human ingenuity and creativity. Furthermore, unlike other problems studied in GP, many different solutions are often possible for a given design problem. For example, there are a multitude of useful designs for an energy-efficient building.******My research will solve increasingly complex design problems by improving the effectiveness of GP in evolutionary design, and will focus on a few major technical issues in evolutionary design. First, complex design problems usually involve a large number of problem-specific considerations, all of which need to be solved if satisfactory solutions are to be obtained. Some of these factors can be very expensive to compute as well, for example, estimating the energy usage of a building using energy simulation software. I will explore ways to effectively handle the many design goals necessary when solving real-world problems. I will also be using the high-performance computing power found in graphics cards, to help speed up the performance of these expensive computations. This will enable enormous increases in processing power, which will make the evolutionary design system much more efficient and powerful. This improvement in computer speed will permit the creation of software support tools that allow users to work interactively with the GP to find suitable designs. A second direction of research will involve creating new design languages to be used by genetic programming. These languages will describe many of the domain-specific design goals for specific problems, and improve the problem-solving abilities of GP. By tackling these two issues, the automatic creation of solutions to complex, real-world design problems will be possible. During the course of the research, I will explore new design problems throughout 2D graphics for art applications, 3D modelling, energy efficient architecture, and biological modelling. These improvements to GP will also be widely applicable to problem areas outside of evolutionary design. A long-term goal is to use this technology to create more intelligent support software for end-users.**