Automatic Construction of State Evaluation Functions and its Dynamic Evaluator

状态评价函数的自动构造及其动态评价器

• 批准号: 13680434
• 负责人: KAWAI Satoru
• 金额: $ 2.3万
• 依托单位: The University of Tokyo
• 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 2002
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-13680434/
• 关键词: evaluation functions; dynamic evaluator; knowledge acquisition; state transition rule; feature; Galoi lattice; reflective computing; ガロア束

The main target of this research is a systematic treatment of the three components; state evaluation by learning, concept formation, and the automatic construction of efficient evaluators. In the first year of the research, a set of evaluation elements (features) with reasonable calculation efficiency is successfully constructed. A thorough analysis of the features, extraction of useful concepts, and a research of making realistic evaluators are performed in the following (and the final) year. Main results follow.・・Advanced research with respect to evaluation functionsIn the field of the automatic construction of evaluation functions, both rapid generation of large sets of functions and efficient processing of these functions are inevitable, beside the theoretical clearness of generation and processing. We show some satisfactory results for these topics with the use of evaluation networks and geometrical patterns.・・Preparatory works for the framework of concept formationWorks in the field of semistructure data is carried out for the structuring of evaluation features. Some mathematical methods of automatic categorization of concepts are investigated. A process algebra which allows the concept of simultaneous events is proposed, with a clear mathematical soundness.・・Construction of runnable evaluatorEach evaluator corresponding to evaluation features are to be implemented, in our research, in Java language by the use of specialization in the level of byte code programs. Some reflective methods for this implementation are investigated.The results of our research are believed to show the applicability to practical problems in the fields of, for example, money market and data mining, not restricted to 2-players games which we used as a workout example.

本研究的主要目标是一个系统的治疗的三个组成部分，学习状态评估，概念形成，和自动建设的有效评估。在第一年的研究中，成功地构建了一套具有合理计算效率的评价元素（特征）。在接下来的一年（也是最后一年），对这些特征进行了彻底的分析，提取了有用的概念，并进行了现实评估者的研究。主要结果如下：··在评价函数的自动构造领域，除了理论上的清晰之外，快速生成大量的函数集和有效地处理这些函数是不可避免的。我们使用评估网络和几何模式为这些主题展示了一些令人满意的结果。··准备工作的框架的概念formationWorks领域的半结构化数据进行结构化的评价功能。研究了概念自动分类的一些数学方法。提出了一种允许同时事件概念的进程代数，具有明确的数学合理性。··构造可运行的求值器每个求值器对应的求值功能都要实现，在我们的研究中，在Java语言中使用字节码程序的层次专门化。我们的研究结果被认为是显示的实际问题，例如，货币市场和数据挖掘领域的适用性，而不限于2人游戏，我们作为一个锻炼的例子。

项目成果

Matsuda, Y.: "The Info Min Principle : A Unifying Information-Based Criterion for Forming Topological Mappings"Proc. ICONIP. 14-19 (2001)

Matsuda, Y.：“信息最小原则：形成拓扑映射的统一的基于信息的标准”Proc。

Kaneko, T.: "Automatic Construction of Pattern-Based Evaluation Functions for Game Programming"IPSJ J. 43(W). 3040-3047 (2002)

Kaneko, T.：“游戏编程中基于模式的评估函数的自动构建”IPSJ J. 43(W)。

Kaneko, T, Yamaguchi, K., and Kawai, S.: "Automatic Feature Construction and Optimization for General Game Player"Proc.6^<th> Game Programming Workshop, IPSJ. 25-32 (2001)

Kaneko, T、Yamaguchi, K. 和 Kawai, S.：“一般游戏玩家的自动特征构建和优化”Proc.6^<th> 游戏编程研讨会，IPSJ。

Matsuda, Y.: "The InfoMin Principle: A Unifying Information-Based Criterion for Forming Topological Mappings"Proc. ICONIP. 14-19 (2001)

Matsuda, Y.：“InfoMin 原理：形成拓扑映射的基于信息的统一标准”Proc。

金子知適: "述語論理で表された特徴関数のネットワーク型局面評価器への自動変換"電子情報通信学会論文誌. J85D1・11. 1074-1082 (2002)

Tomoyuki Kaneko：“将谓词逻辑表示的特征函数自动转换为网络型情况评估器”，电子信息通信工程师学会汇刊 J85D1·11（2002 年）。