人工知能学会論文誌
検索
OR
閲覧
検索
18 巻 , 5 号
Technical Papers
選択された号の論文の8件中1~8を表示しています
    • |<
    • <
    • 1
    • >
    • >|
論文
  • 弘中 大介, 横田 将生
    原稿種別: 論文
    専門分野: 自然言語
    18 巻 (2003) 5 号 p. 233-244
    公開日: 2003/06/10
    ジャーナル フリー
    In general, it is not always easy for people to communicate each other comprehensively by limited information media. In such a case, employment of another information medium is very helpful and therefore cross-media translation is very important during such a communication. This paper presents the method and experiment of cross-media translation based on MIDST(Mental Image Directed Semantic Theory), where natural language texts about static positional relations of physical objects are systematically interpreted into 2-D pictures.
    抄録全体を表示
  • 柏原 昭博, 坂本 雅直, 長谷川 忍, 豊田 順一
    原稿種別: 論文
    専門分野: その他
    18 巻 (2003) 5 号 p. 245-256
    公開日: 2003/06/12
    ジャーナル フリー
    Hypermedia/hypertext-based resources for learning generally provide learners with hyperspace, which consists of pages and links among the pages. In the hyperspace, they can navigate the pages in a self-directed way to learn the domain concepts/knowledge. The navigation often involves constructing knowledge, in which they would make semantic relationships among the contents learned in the navigated pages. Such self-directed learning in hyperspace requires learners to reflect on their knowledge construction process, which they have carried out so far, since what and how they have learned becomes hazy as the navigation progresses. However, it is hard for them to keep reflection during navigating hyperspace. The main issue addressed in this paper is how to facilitate learners' reflection to promote their self-directed learning. Our approach to this issue is to provide learners with a learning tool, which allows learners to annotate their navigation history representation with their knowledge construction process. The annotated navigation history enables them to reflect on their knowledge construction process. This paper also demonstrates an interactive history, which generates the annotated navigation history from learners' annotation. It also generates a knowledge map that visualizes the semantic relationships among the pages learners have learned in hyperspace. This paper also describes a case study with the interactive history. The results indicate that it facilitates reflection on knowledge construction process carried out in hyperspace.
    抄録全体を表示
  • 西村 芳男, 鷲尾 隆, 吉田 哲也, 元田 浩, 猪口 明博, 岡田 孝
    原稿種別: 論文
    専門分野: 学習と発見
    18 巻 (2003) 5 号 p. 257-268
    公開日: 2003/06/24
    ジャーナル フリー
    Apriori-based Graph Mining (AGM) algorithm efficiently extracts all the subgraph patterns which frequently appear in graph structured data. The algorithm can deal with general graph structured data with multiple labels of vartices and edges, and is capable of analyzing the topological structure of graphs. In this paper, we propose a new method to analyze graph structured data for a 3-dimensional coordinate by AGM. In this method the distance between each vertex of a graph is calculated and added to the edge label so that AGM can handle 3-dimensional graph structured data. One problem in our approach is that the number of edge labels increases, which results in the increase of computational time to extract subgraph patterns. To alleviate this problem, we also propose a faster algorithm of AGM by adding an extra constraint to reduce the number of generated candidates for seeking frequent subgraphs. Chemical compounds with dopamine antagonist in MDDR database were analyzed by AGM to characterize their 3-dimensional chemical structure and correlation with physiological activity.
    抄録全体を表示
  • 松原 繁夫
    原稿種別: 論文
    専門分野: エージェント・分散人工知能
    18 巻 (2003) 5 号 p. 269-277
    公開日: 2003/07/15
    ジャーナル フリー
    This paper focuses on a task allocation problem, especially cases where the task is to find a solution in a search problem or a constraint satisfaction problem. If the search problem is hard to solve, a contractor may fail to find a solution. Here, the more computational resources such as the CPU time the contractor invests in solving the search problem, the more a solution is likely to be found. This brings about a new problem that a contractee has to find an appropriate level of the quality in a task achievement as well as to find an efficient allocation of a task among contractors. For example, if the contractee asks the contractor to find a solution with certainty, the payment from the contractee to the contractor may exceed the contractee's benefit from obtaining a solution, which discourages the contractee from trading a task. However, solving this problem is difficult because the contractee cannot ascertain the contractor's problem-solving ability such as the amount of available resources and knowledge (e.g. algorithms, heuristics) or monitor what amount of resources are actually invested in solving the allocated task. To solve this problem, we propose a task allocation mechanism that is able to choose an appropriate level of the quality in a task achievement and prove that this mechanism guarantees that each contractor reveals its true information. Moreover, we show that our mechanism can increase the contractee's utility compared with a simple auction mechanism by using computer simulation.
    抄録全体を表示
  • 村上 陽平, 石田 亨, 河添 智幸, 菱山 玲子
    原稿種別: 論文
    専門分野: エージェント・分散人工知能
    18 巻 (2003) 5 号 p. 278-285
    公開日: 2003/07/15
    ジャーナル フリー
    Making it easier to design interactions between agents and humans is essential for realizing multi-agent simulations of social phenomena such as group dynamics. To realize large-scale social simulations, we have developed the scenario description languages Q and IPC (Interaction Pattern Card); they enable experts in the application domain (often not computing professionals) to easily create complex scenarios. We have also established a four-step process for creating scenarios: 1) defining a vocabulary, 2) describing scenarios, 3) extracting interaction patterns, and 4) integrating real and virtual experiments. In order to validate the scenario description languages and the four-step process, we ran a series of evacuation simulations based on the proposed languages and process. We successfully double-check the result of the previous controlled experiment done in a real environment.
    抄録全体を表示
  • 宮崎 和光, 小林 重信
    原稿種別: 論文
    専門分野: 創発システム
    18 巻 (2003) 5 号 p. 286-296
    公開日: 2003/07/17
    ジャーナル フリー
    We know the rationality theorem of Profit Sharing(PS) [Miyazaki 94, Miyazaki 99b] and the Rational Policy Making algorithm(RPM) [Miyazaki 99a] to guarantee the rationality in a typical class of Partially Observable Markov Decision Processes (POMDPs). In this paper, we focus on the whole class of POMDPs and propose PS-r that is an algorithm connected PS and RPM with random selection. In the first, we have analyzed the behavior of PS-r. We have derived that the maximum value of the step to get a reward by PS-r divided by that of random selection is
    $({\Large r\frac{(1+\frac{M-1}{r})^n}{M^n}})$ where $n$
    is the maximum number of state that senses same state due to the agent's sensory limitation and M is the number of actions. Furthermore, we propose PS-r* that can improve the behavior of PS-r. Through numerical examples, we conform the effectiveness of PS-r*.
    抄録全体を表示
  • 坂東 秀之, 井上 克巳, 鍋島 英知
    原稿種別: 論文
    専門分野: 基礎・理論
    18 巻 (2003) 5 号 p. 297-304
    公開日: 2003/07/17
    ジャーナル フリー
    Recent work on representing action and change has introduced high-level action languages which describe the effects of actions as causal laws in a declarative way. Among such action languages, the language A is the first and the most basic language. In dynamic domains, an agent needs the ability to react against changes of environment and to generate robust plans for a long-term goal, and appropriate representation is necessary for this purpose. In real problems, however, it is difficult to describe complete causal laws for the domain, but it is easier to get observations. In this paper, we propose an algorithm to learn causal laws from an incomplete domain description in the language A, given observations after performing action sequences. Our learning algorithm generates causal laws based on an algorithm to learn finite automata. We also prove the correctness of the learning algorithm. From the result in this work, induction of the effects of actions can now be formally characterized within action languages.
    抄録全体を表示
  • 安藤 晋, 伊庭 斉志
    原稿種別: 論文
    専門分野: 創発システム
    18 巻 (2003) 5 号 p. 305-315
    公開日: 2003/07/29
    ジャーナル フリー
    Evolutionary computation has been applied to numerous design tasks, including design of electric circuits, neural networks, and genetic circuits. Though it is a very effective solution for optimizing network structures, genetic algorithm faces many difficulties, often referred to as the permutation problems, when both topologies and the weights of the network are the target of optimization. We propose a new crossover method used in conjunction with a genotype with information tags. The information tags allow GA to recognize and preserve the common structure of parent chromosomes during genetic crossover. The method is implemented along with subpopulating strategies to make the parallel evolution of network topology and weights feasible and efficient. The proposed method is evaluated on a few typical and practical problems, and shows improvement from conventional methodologies and genotypes.
    抄録全体を表示
    • |<
    • <
    • 1
    • >
    • >|
feedback
Top