Acquisition of Evaluation Function of high precision by Learning

通过学习获得高精度评价函数

• 批准号: 15500084
• 负责人: KAWAI Satoru
• 金额: $ 2.5万
• 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: 2003
• 资助国家: 日本
• 起止时间: 2003 至 2004
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-15500084/
• 关键词: evaluation functions; dynamic evaluator; knowledge acquisition; state transition rules; Galois lattice; reflective calculation; non-wellfounded set; aspect; feature

The main and final target of this research is the automatic construction of evaluators by learning, integrating state evaluation by learning, concept formation, and creation of efficient evaluators. Main results of the component researches follow.1.Automatic construction of evaluation functionsThe evaluation functions are constructed, without human preparation, only from the basic rules and framework of the target field. The essential part of this construction is the automatic creation of simple data sets for state evaluation, followed by the selection based on the appropriateness of the data. Following some preparatory studies carried out in 2001 and 2002, more precise evaluation functions are obtained by the analysis of bulk data for states evaluation.2.Concept formation and structure analysisThe structure of a Galois lattice, based on a partial order defined among the set of simple piecewise statuses, is studied as a framework of concepts. In this research, semistructured data, frequently found in the real world, is also studied. The "summary structure" can be extracted from such "non-well ordered structure" by an analytical method.3.Reflection and aspectsIn the target system of this research, a lot of software agents for state evaluation are to be prepared by the method of reflection. These agents are expected to act optimally in the real evaluation referring the concept database. In this research, the framework of aspect oriented programming is studied for interclass function increments. Some results in support systems and theoretical framework in aspect oriented programming are obtained.

本研究的主要和最终目标是通过学习自动构建评估器，整合学习状态评估、概念形成和高效评估器的创建。各部分研究的主要成果如下： 1.评价函数的自动构建 评价函数的构建无需人工准备，仅从目标领域的基本规则和框架出发。这种构建的本质部分是自动创建用于状态评估的简单数据集，然后根据数据的适当性进行选择。经过2001年和2002年进行的一些准备性研究，通过对状态评估的大量数据进行分析，获得了更精确的评估函数。2.概念形成和结构分析伽罗瓦格的结构，基于在集合中定义的偏序。简单的分段状态，作为概念框架进行研究。在这项研究中，还研究了现实世界中常见的半结构化数据。通过解析的方法可以从这种“非良序结构”中提取出“概要结构”。 3.反思与方面 在本研究的目标系统中，将采用以下方法来编写大量的状态评估软件代理。反射。这些代理预计在参考概念数据库的实际评估中表现最佳。在本研究中，针对类间函数增量，研究了面向方面编程的框架。在面向方面编程的支持系统和理论框架方面取得了一些成果。

项目成果

Evaluation Functions Based on Pairs of Pieces

基于棋子对的评估函数

• 发表时间: 2003
• 期刊: The 8^<th> Game Programming Workshop IPSJ
• 作者: Kaneko;T.;Tanaka;T.;Yamaguchi;K.;Kawai;S.
• 通讯作者: S.

田中 哲朗: "高速将棋ライブラリOpenShogiLibの作成"第8回ゲームプログラミングワークショップ. 8. 114-117 (2003)

Tetsuro Tanaka：“高速将棋库 OpenShogiLib 的创建”第 8 届游戏编程研讨会 8. 114-117 (2003)。

Structural Analysis for Web Documentation Using the Non-Well-Founded Set

使用非良好基础集对 Web 文档进行结构分析

• 发表时间: 2004
• 期刊: Fifteenth ACM Conference on Hypertext and Hypermedia 15
• 作者: Horie;I
• 通讯作者: I

Kaneko, T.: "Automated Identification of Patterns in Evaluation Functions for General Game Players"Advances in Computer Games. 10. 279-298 (2003)

Kaneko, T.：“一般游戏玩家评估函数模式的自动识别”计算机游戏的进展。

A Compilation and Optimization Model for Aspect-Oriented Programs

• DOI: 10.1007/3-540-36579-6_4
• 发表时间: 2003-04
• 作者: Hidehiko Masuhara;G. Kiczales;Christopher Dutchyn