Emergent Design Theories in Interactive Information Fields

该项目旨在阐明和诠释在设计人工制品的过程中所揭示的人的创造力和综合能力，并包括以下分主题：(1)分析了描述人与计算机之间信息流的“交互信息场”的数学结构；(2)基于学习的紧急设计理论（将强化学习应用于环境信息不完整的问题）；(3)基于交互的紧急设计理论（将交互进化计算应用于规范不完整的问题）。在子主题(1)的过程中，首先以分类器系统为例，研究了基于情境理论和通道理论的分类器系统的数学建模。在抽象设计理论方面，以溯因论为重点，给出了非演绎推理的数学表述，阐明了归纳、溯因论和系统设计之间的关系。在子主题(2)中，将基于强化学习和基于遗传的机器学习框架重构为获取环境内部模型并寻求解决方案的机制。通过对实时调度、两足机器人行走模式生成、机器人导航等问题的计算实验，验证了该框架的有效性。此外，在子主题(3)中，主要考虑了利用人类模糊决策的交互式进化计算方法。研究了协同进化遗传算法和带性选择遗传算法的计算模型，并通过计算实验验证了这些模型的可行性。

The project aimed at clarifying and also interpreting the creativity and the comprehensive ability of human which were revealed in the process of designing artifacts, and included the following sub-themes : (1) analysis of the mathematical structure of "interactive information fields" which describe the informational flow between human and computers, (2) emergent design theories based on learning (application of reinforcement learning to such problems that the information on an environment is incomplete), and (3) emergent design theories based on interaction (application of interactive evolutionary computation to such problems that the specification is also incomplete).In the course of the sub-theme (1), first by taking classifier systems for the example, their mathematical modeling based on the situation theory and the channel theory was investigated. As for the abstract design theory, by focusing on abduction, the mathematical formulation of non-deductive reasoning was given, which clarified the relationship among induction, abduction and system design.In the sub-theme (2), the frameworks based on reinforcement learning and genetics-based machine learning were reconstructed as the mechanism of acquiring the internal model of an environment as well as seeking for a solution. Through some computational experiments for the problems of real-time scheduling, generating a walking pattern of biped robots, navigating a robot, and so on, the effectiveness of the frameworks was examined.Furthermore, in the sub-theme (3), methods of interactive evolutionary computation, which utilized the ambiguous decisions of human, were mainly considered. The computational models of a co-evolutionary genetic algorithm as well as a genetic algorithm with sexual selection were investigated, and the potential of these models was confirmed through computational experiments.