人工知能学会論文誌
検索
OR
閲覧
検索
21 巻 , 3 号
選択された号の論文の11件中1~11を表示しています
    • |<
    • <
    • 1
    • >
    • >|
一般論文
論文
  • 元木 達也, 沼口 靖
    21 巻 (2006) 3 号 p. 219-230
    公開日: 2006/02/03
    ジャーナル フリー
    This paper is motivated by an experimental result that better performing genetic programming runs tend to have higher phenotypic diversity. To maintain phenotypic diversity, we apply implicit fitness sharing and its variant, called unfitness multiplying. To apply these methods to problems in which individuals have infinite kinds of possible behaviours, we classify posible behaviours into 50 achievement levels, and assign a reward or a penalty to each level. In implicit fitness sharing a reward is shared out among individuals with the same achievement level, and in unfitness multiplying a penalty is multiplied by the number of individuals with the same level and is distributed to related individuals. Five benchmark problems (11-multiplexer, sextic polynomial, four-sine, intertwined spiral, and artificial ant problems) are used to illustrate the effect of the methods. The results show that our methods clearly promote diversity and lead population to a smooth frequency distribution of achievement levels, and that our methods usually perform better than the original implicit fitness sharing on success rate and the best (raw) fitness. We also observe that the unfitness multiplying makes a quite different ranking over individuals than the one by the implicit fitness sharing.
    抄録全体を表示
  • Sasiporn Usanavasin, Shingo Takada, Norihisa Doi
    21 巻 (2006) 3 号 p. 231-242
    公開日: 2006/02/03
    ジャーナル フリー
    The Semantic Web and ontologies can be seen as a way to enable greater access not only to contents but also services on the Web. Users and software agents should be able to discover, invoke and compose Web services with a high degree of automation. However, current existing Web services are created by many different parties and are described based on different ontologies, which makes it difficult for retrieval systems to locate needed services. To facilitate the discovery of Web services in multi-ontology environments, we propose an approach to determine the semantic similarity of properties between different ontologies.
    抄録全体を表示
  • 村上 陽平, 杉本 悠樹, 石田 亨
    21 巻 (2006) 3 号 p. 243-250
    公開日: 2006/02/03
    ジャーナル フリー
    Constructing highly realistic agents is essential if agents are to be employed in virtual training systems. In training for collaboration based on face-to-face interaction, the generation of emotional expressions is one key. In training for guidance based on one-to-many interaction such as direction giving for evacuations, emotional expressions must be supplemented by diverse agent behaviors to make the training realistic. To reproduce diverse behavior, we characterize agents by using a various combinations of operation rules instantiated by the user operating the agent. To accomplish this goal, we introduce a user modeling method based on participatory simulations. These simulations enable us to acquire information observed by each user in the simulation and the operating history. Using these data and the domain knowledge including known operation rules, we can generate an explanation for each behavior. Moreover, the application of hypothetical reasoning, which offers consistent selection of hypotheses, to the generation of explanations allows us to use otherwise incompatible operation rules as domain knowledge. In order to validate the proposed modeling method, we apply it to the acquisition of an evacuee's model in a fire-drill experiment. We successfully acquire a subject's model corresponding to the results of an interview with the subject.
    抄録全体を表示
ショートノート
論文
  • 吉岡 卓, 東条 敏
    21 巻 (2006) 3 号 p. 257-265
    公開日: 2006/03/09
    ジャーナル フリー
    Linear tense logics are widely accepted for structural temporal representation, where the basic KT has two modal operators G and H, each of which represents the future and the past, respectively. On the other hand, the temporal interval relations arranged by Allen have long been the standard of natural language semantics, though it still lacks the modal-logical foundation. Van Benthem proposed ∉up and ∉down in regard to the accessibility to overlapping intervals and subintervals, respectively; however, the logical feature of the modality has not well studied. In this study, we propose a many-dimensional logic including the conventional tense logic, together with such interval accessibility. And, we show that our logic provide a formal apparatus for a precise aspectual classification. Lastly, we introduce the sequent system for our logic. We show the subformula property holds in our system, and thus would be able to show the decidability.
    抄録全体を表示
  • 水澤 雅高, 栗原 正仁
    21 巻 (2006) 3 号 p. 266-275
    公開日: 2006/03/09
    ジャーナル フリー
      Although the maze (or gridworld) is one of the most widely used benchmark problems for real-time search algorithms, it is not sufficiently clear how the difference in the density of randomly positioned obstacles affects the structure of the state spaces and the performance of the algorithms. In particular, recent studies of the so-called phase transition phenomena that could cause dramatic change in their performance in a relatively small parameter range suggest that we should evaluate the performance in a parametric way with the parameter range wide enough to cover potential transition areas.
    In this paper, we present two measures for characterizing the hardness of randomly generated mazes parameterized by obstacle ratio and relate them to the performance of real-time search algorithms. The first measure is the entropy calculated from the probability of existence of solutions. The second is a measure based on total initial heuristic error between the actual cost and its heuristic estimation. We show that the maze problems are the most complicated in both measures when the obstacle ratio is around 41\%. We then solve the parameterized maze problems with the well-known real-time search algorithms RTA*, LRTA*, and MARTA* to relate their performance to the proposed measures. Evaluating the number of steps required for a single problem solving by the three algorithms and the number of those required for the convergence of the learning process in LRTA*, we show that they all have a peak when the obstacle ratio is around 41\%. The results support the relevance of the proposed measures. We also discuss the performance of the algorithms in terms of other statistical measures to get a quantitative, deeper understanding of their behavior.
    抄録全体を表示
  • 山本 浩司, 片上 大輔, 新田 克己, 相場 亮, 桑田 仁
    21 巻 (2006) 3 号 p. 276-286
    公開日: 2006/03/17
    ジャーナル フリー
    In this paper, we propose a method for estimating the credibility of the posted information from users. We incorporate this method in a system which recommends the route and destination using other user's posted information. Users can post information, and other users can refer to them. These information includes a picture, comment, genre, expiration date, and so on. The system displays these information on the map. Since posted information can include subjective information from various perspectives, we can't trust all of the postings as they are. We propose and integrate two factors of credibility of posted information. First, credibility of voting by other users is determined based on the posting user's own credibility and other users' credibility who approved or disapproved her/his posting information. The other is credibility of geographic posting tendency of the users. This credibility is determined based on dispersion of posted information on the map. From experimental results, we concluded our method measures credibility of posted information validly.
    抄録全体を表示
  • 鳥居 大祐, 石田 亨, François Bousquet
    21 巻 (2006) 3 号 p. 287-294
    公開日: 2006/03/17
    ジャーナル フリー
    In a participatory approach by social scientists, role playing games (RPG) are effectively used to understand real thinking and behavior of stakeholders, but RPG is not sufficient to handle a dynamic process like negotiation. In this study, a participatory simulation where user-controlled avatars and autonomous agents coexist is introduced to the participatory approach for modeling negotiation. To establish a modeling methodology of negotiation, we have tackled the following two issues. First, for enabling domain experts to concentrate interaction design for participatory simulation, we have adopted the architecture in which an interaction layer controls agents and have defined three types of interaction descriptions (interaction protocol, interaction scenario and avatar control scenario) to be described. Second, for enabling domain experts and stakeholders to capitalize on participatory simulation, we have established a four-step process for acquiring negotiation model: 1) surveys and interviews to stakeholders, 2) RPG, 3) interaction design, and 4) participatory simulation. Finally, we discussed our methodology through a case study of agricultural economics in the northeast Thailand.
    抄録全体を表示
  • 大谷 紀子, 貝原 巳樹雄, 志村 正道
    21 巻 (2006) 3 号 p. 295-300
    公開日: 2006/03/17
    ジャーナル フリー
    This paper proposes a novel method for generating a decision tree to discriminate polymers accurately with the near-infrared rays spectrum. The polymer discrimination system is needed for recycling plastics, and the near-infrared rays spectrum is useful for rapid and non-destructive discrimination. The former system SESAT, which is based on symbiotic evolution, can generate simple and accurate trees, but is not effective for data that has a lot of attributes like the near-infrared rays spectrum. We design the structure of the partial solution ``sprig'' for sufficient learning, and the fitness function of the whole solution ``decision tree blueprint'' for 2-class discrimination. In addition, we introduce two-step discrimination with the aim of obtaining higher accuracy. In the first step, examples are divided into two groups, one group being easier than the other to discriminate by a tree. In the second step, two trees are generated that discriminate one kind of polymer from the others, for two groups of examples. By doing this, a minority of examples is also discriminated accurately. Based on this method we developed a polymer discrimination system called TS-SEPT. Our experimental results on real data of polymers show that the accuracy of TS-SEPT compares favorably with that of the other systems, the similar system without two-step discrimination, SESAT and C5.0. It emerged that both the method for generating decision trees and two-step discrimination contributed to the improved accuracy.
    抄録全体を表示
  • 藤野 昭典, 上田 修功, 斉藤 和巳
    21 巻 (2006) 3 号 p. 301-309
    公開日: 2006/03/23
    ジャーナル フリー
    Semi-supervised classifier design that simultaneously utilizes both a small number of labeled samples and a large number of unlabeled samples is a major research issue in machine learning. Existing semi-supervised learning methods for probabilistic classifiers belong to either generative or discriminative approaches. This paper focuses on a semi-supervised probabilistic classifier design for multiclass and single-labeled classification problems and first presents a hybrid approach to take advantage of the generative and discriminative approaches. Our formulation considers a generative model trained on labeled samples and a newly introduced bias correction model, whose belongs to the same model family as the generative model, but whose parameters are different from the generative model. A hybrid classifier is constructed by combining both the generative and bias correction models based on the maximum entropy principle, where the combination weights of these models are determined so that the class labels of labeled samples are as correctly predicted as possible. We apply the hybrid approach to text classification problems by employing naive Bayes as the generative and bias correction models. In our experimental results on three English and one Japanese text data sets, we confirmed that the hybrid classifier significantly outperformed conventional probabilistic generative and discriminative classifiers when the classification performance of the generative classifier was comparable to the discriminative classifier.
    抄録全体を表示
技術資料
  • 熊本 忠彦, 太田 公子
    21 巻 (2006) 3 号 p. 310-318
    公開日: 2006/03/28
    ジャーナル フリー
    Impression-based music retrieval helps users in finding musical pieces that suit their preferences, feelings, or mental states from the huge volume of a music database. We have therefore developed an impression-based music retrieval system that enables this. Users are asked to select one or more pairs of impression words from the multiple pairs presented by the system and estimate each of the selected pairs on a seven-step scale in order to input their impressions into the system. For instance, if they want to locate musical pieces that will create a happy impression, they should check the radio button ``Happy'' in the impression scale, ``Very happy -- Happy -- A little happy -- Neutral -- A little sad -- Sad -- Very sad,'' where a pair of impression words with a seven-step scale is called an ``impression scale'' in this paper. The system would measure the distance between the impressions of every musical piece in a user-specified music database and the impressions inputted by the user, and determine candidate musical pieces to be presented as retrieval results. In this paper, we define the form of vectors that numerically express impressions of musical pieces, and propose a method of generating such a vector from a musical piece. The most significant attribute of this method is that it uses n-gram statistics of information on pitch, strength, and length of every tone in that musical piece as features extracted from it. We also present the results of evaluating the performance of the system.
    抄録全体を表示
    • |<
    • <
    • 1
    • >
    • >|
feedback
Top