Interactive Erolutionary Computation with Visualization of an oprimization Landscape

具有优化景观可视化的交互式演化计算

基本信息

批准号：
13680451
负责人：
TAKAGI Hideyuki
金额：
$ 1.73万
依托单位：
Kyushu University(Faculty of Design)
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
2001
资助国家：
日本
起止时间：
2001 至 2003
项目状态：
已结题

项目摘要

This research consists of three parts.The first part is the proposal of visualized interactive evolutionary computation (Visualized IEC) that visualizes the landscape of an EC search space and the evaluation of its effectiveness. Human IEC user and EC play independent roles of evaluation and optimization in a basic IEC approach, respectively. We proposed a method that maps an EC landscape into a 2-D space for visualizing the landscape, displays the visualized EC landscape to a human user, and let the user join to the EC optimization process. We compared four mapping methods from the computational cost and convergence speed point of view and concluded that the Visualized IEC with SOM (Self-Organized Map) is the best practical implementation among the four mapping methods.The second part is the development of this visualization approach. First, we implemented the Visualized IEC into a PDA for reducing the restrictions of place and time of hearing-aid fitting and evaluated the practicalne … More ss of the EC-based hearing-aid fitting method. Second, we proposed an acceleration method of EC convergence by using EC landscape information. We evaluated the method that estimates a global optimum area by approximating an EC landscape using a single-peak function and used the estimated global optimum as an elite individual in the next generation, the method that approximates an EC landscape with multiple Gaussian functions as well and generates offspring by using the functions as random Gaussian noise generators, and other methods. Third, we applied an IEC technique to design medical image enhancement filters visually for increasing the intelligibility of the images for medical diagnostics. We showed its effectiveness by using a medical image diagnostic expert who is not an expert of image signal processing as an IEC experimental subject.The third part is the investigation of the IEC applicability to the visualization of mental condition. We had 3 schizophrenics and 5 mental-normals design 'happy' and 'sad' computer graphics (CG) lightings with an IEC-based CG lighting design support system and had 32 subjects evaluate their designs with a paired comparison method. This experimental result showed that the dynamic expression ranges of "happy -sad" of the schizophrenics were narrower than those of mental-normals significantly. It means that the IEC technique can be a capable tool for measuring human mind situation. Less

本研究由三个部分组成。第一部分是可视化交互式进化计算（Visualized IEC）的提出，该计算可视化 EC 搜索空间的景观并评估其有效性，人类 IEC 用户和 EC 发挥独立的评估和优化作用。我们分别提出了一种将 EC 景观映射到二维空间以可视化景观的方法，向人类用户显示可视化的 EC 景观，并让用户加入到 EC 优化过程中。从计算成本和收敛速度的角度比较了四种映射方法，得出结论：带有SOM（自组织映射）的可视化IEC是四种映射方法中最好的实际实现。第二部分是这种可视化方法的发展。首先，我们将可视化IEC应用到PDA中，以减少助听器验配地点和时间的限制，并评估基于EC的助听器验配方法的实用性。其次，我们提出了一种EC加速方法。我们评估了通过使用单峰函数近似 EC 景观来估计全局最佳区域的方法，并将估计的全局最佳值用作下一代的精英个体，即近似 EC 景观的方法。第三，我们应用 IEC 技术来视觉设计医学图像增强滤波器，以提高图像的清晰度。我们通过使用非图像信号处理专家的医学图像诊断专家作为IEC实验对象来展示其有效性。第三部分是IEC对精神状况可视化的适用性研究。精神分裂症患者和 5 名心理正常的人使用基于 IEC 的 CG 照明设计支持系统设计“快乐”和“悲伤”的计算机图形 (CG) 照明，并让 32 名受试者使用配对比较方法评估他们的设计。精神分裂症患者的“快乐-悲伤”动态表达范围明显小于心理正常人，这意味着IEC技术可以成为测量人类心理状况的有力工具。