人工知能学会論文誌
検索
OR
閲覧
検索
22 巻 , 5 号
選択された号の論文の10件中1~10を表示しています
    • |<
    • <
    • 1
    • >
    • >|
一般論文
論文
  • Kevin J. Binkley, Ken Seehart, Masafumi Hagiwara
    22 巻 (2007) 5 号 p. 461-471
    公開日: 2007/06/07
    ジャーナル フリー
    In this study, we use temporal difference learning (TDL) to investigate the ability of 20 different artificial neural network (ANN) architectures to learn othello game board evaluation functions. The ANN evaluation functions are applied to create a strong othello player using only 1-ply search. In addition to comparing many of the ANN architectures seen in the literature, we introduce several new architectures that consider the game board symmetry. Both embedding the game board symmetry into the network architecture through weight sharing and the outright removal of symmetry through symmetry removal are explored. Experiments varying the number of inputs per game board square from one to three, the number of hidden nodes, and number of hidden layers are also performed. We found it advantageous to consider game board symmetry in the form of symmetry by weight sharing; and that an input encoding of three inputs per square outperformed the one input per square encoding that is commonly seen in the literature. Furthermore, architectures with only one hidden layer were strongly outperformed by architectures with multiple hidden layers. A standard weighted-square board heuristic evaluation function from the literature was used to evaluate the quality of the trained ANN othello players. One of the ANN architectures introduced in this study, an ANN implementing weight sharing and consisting of three hidden layers, using only a 1-ply search, outperformed a weighted-square test heuristic player using a 6-ply minimax search.
    抄録全体を表示
  • 若木 利子, 富田 一夫
    22 巻 (2007) 5 号 p. 472-481
    公開日: 2007/06/07
    ジャーナル フリー
    Recently we proposed a method of compiling prioritized circumscription into answer set programming. However, its encoding has the guess and check structure, where the candidates are generated by the guess program and the check is expressed by the inconsistency of the check program. In this paper, we present another method which compiles prioritized circumscription into a single general disjunctive program (GDP) by means of integration of the guess and check programs. The answer sets of the transformed GDP yield models of a given circumscription by emulating the inconsistency of the check program using integrity constraints. Thanks to our integration technique, the circumscriptive theorem prover to evaluate a query with respect to prioritized circumscription is easily established by expressing the inconsistency check in a single GDP. Our experimental results show that the performance of the tool we have implemented using our new method has not only improved remarkably, i.e. reduction of runtime as compared to our previous method, but has also exceeded that of the recently developed software tool, prio_circ2dlp, for prioritized circumscription as far as our experiments are concerned.
    抄録全体を表示
  • 寺内 敦, 明石 修, 丸山 充, 菅原 俊治, 福田 健介, 廣津 登志夫, 栗原 聡, 小柳 惠一
    22 巻 (2007) 5 号 p. 482-492
    公開日: 2007/06/12
    ジャーナル フリー
    An agent organization system for multi-agent based network management, called ARTISTE, is proposed. For a multi-agent system (MAS) to solve problems effectively, it is important to organize agents appropriately. Organizing agents adaptively on the Internet, however, is not easy, because the status of the Internet changes dynamically in a short time and no one can have a complete view of the whole network. The aims of ARTISTE are to form an agent organization in accordance with the current Internet and its problem-solving context and to provide organizational information for target MASs. ARTISTE operates as an independent system with respect to any MAS. To organize agents, ARTISTE collects information about agents' abilities and statuses, network information such as topologies, and a problem-dependent requirement from a target MAS. ARTISTE is designed as an MAS, and it can collect the information about the network and the target MAS from multiple observation points. Furthermore, ARTISTE agents exchange their own local information in order to create a more global view of the network and the distribution of the agents. Our performance evaluation showed that ARTISTE works with sufficient performance for use in the actual Internet environment.
    抄録全体を表示
  • 岡田 将吾, 賀 小淵, 小島 量, 長谷川 修
    22 巻 (2007) 5 号 p. 493-507
    公開日: 2007/06/19
    ジャーナル フリー
    This paper presents an unsupervised approach of integrating speech and visual information without using any prepared data(training data). The approach enables a humanoid robot, Incremental Knowledge Robot 1 (IKR1), to learn words' meanings. The approach is different from most existing approaches in that the robot learns online from audio-visual input, rather than from stationary data provided in advance. In addition, the robot is capable of incremental learning, which is considered to be indispensable to lifelong learning. A noise-robust self-organized incremental neural network(SOINN) is developed to represent the topological structure of unsupervised online data. We are also developing an active learning mechanism, called ``desire for knowledge'', to let the robot select the object for which it possesses the least information for subsequent learning. Experimental results show that the approach raises the efficiency of the learning process. Based on audio and visual data, we construct a mental model for the robot, which forms a basis for constructing IKR1's inner world and builds a bridge connecting the learned concepts with current and past scenes.
    抄録全体を表示
  • 生田目 慎也, 伊庭 斉志
    22 巻 (2007) 5 号 p. 508-519
    公開日: 2007/07/04
    ジャーナル フリー
    In order to estimate Gene Regulatory Networks (GRNs) from gene expression time series data, various recurrence or differential equation based models have been proposed, such as S-system, Linear model etc. Generally, it is assumed that a specific recurrence or differential equation model is sufficient to estimate the network from the expression profile. However, with so many different models available, it is not easy to recognize the model that will be most suitable for a particular network inference problem. To deal with the problem, integrative estimation with multiple recurrence or differential equation based models seems promising. In this paper, we propose the integration of multiple estimation methods by means of AdaBoost. Empirical studies show the effectiveness of our proposal.
    抄録全体を表示
  • 佐久間 淳, 小林 重信
    22 巻 (2007) 5 号 p. 520-530
    公開日: 2007/07/06
    ジャーナル フリー
    This paper presents a kernel density estimation method by means of real-coded crossovers. Functions of real-coded crossover operators are composed of probabilistic density estimation from parental populations and sampling from estimated models. Real-coded Genetic Algorithm (RCGA) does not explicitly estimate probabilistic distributions, however, probabilistic model estimation is implicitly included in algorithms of real-coded crossovers. Based on this understanding, we exploit the implicit estimation of probabilistic distribution of crossovers as a kernel density estimator. We also propose an application of crossover kernels to Expectation-Maximization estimation (EM) of Gaussian mixtures.
    抄録全体を表示
  • 松尾 豊, 安田 雪
    22 巻 (2007) 5 号 p. 531-541
    公開日: 2007/07/17
    ジャーナル フリー
    Our purpose here is to (1) investigate the structure of the personal networks developed on mixi, a Japanese social networking service (SNS), and (2) to consider the governing mechanism which guides participants of a SNS to form an aggregate network. Our findings are as follows:the clustering coefficient of the network is as high as 0.33 while the characteristic path lenght is as low as 5.5. A network among central users (over 300 edges) consist of two cliques, which seems to be very fragile. Community-affiliation network suggests there are several easy-entry communities which later lead users to more high-entry, unique-theme communities. The analysis on connectedness within a community reveals the importance of real-world interaction. Lastly, we depict a probable image of the entire ecology on {\\em mixi} among users and communities, which contributes broadly to social systems on the Web.
    抄録全体を表示
  • 永田 裕一
    22 巻 (2007) 5 号 p. 542-552
    公開日: 2007/07/17
    ジャーナル フリー
    We propose an genetic algorithm (GA) that applies to the traveling salesman problem (TSP). The GA uses edge assembly crossover (EAX), which is known to be effective for solving the TSP. We first propose a fast implementation of a localized EAX where localized edge exchanges are used in the EAX procedure. We also propose a selection model with an effective combination of the localized EAX that can maintain population diversity at negligible computational costs. Edge entropy measure is used to evaluate population diversity. We demonstrate that the proposed GA is comparable to state-of-the-art heuristics for the TSP. Especially, the GA is superior to them on large instances more than 10,000 cities. For example, the GA found an optimal solution of brd14051 (14,051 cities instance) with a reasonable computational cost. The results are quite impressive because the GA does not use Lin-Kernighan local search (LKLS) even though almost all existing state-of-the-art heuristics for the TSP based on LKLS and its variants.
    抄録全体を表示
論文特集:近未来チャレンジ
論文
  • 加藤 恒昭, 松下 光範
    22 巻 (2007) 5 号 p. 553-562
    公開日: 2007/07/17
    ジャーナル フリー
    Information compilation is a novel technology that allows it to compile various information intelligently, and to make it easy to understand and to access. In this paper, as an instance of the possibilities of information compilation, we show a framework that extracts and visualizes given time-series information and its changes, and provides users with a multi-modal summarization and also an interactive interface for accessing that information. It can meet information requests, in which users need to comprehend some trend and movement, and access a series of documents containing specific time-series information related. We emphasis the importance of changes of data during some time period rather than data points, as the unit of information extracted and represented. Based on this idea, we propose a visualization method in which qualitative and quantitative characteristics of changes of a given time-series information are plotted with textually represented comments, and a widely applicable information extraction method that regards the changes of time-series information as information primitives and extracts those for the visualization.
    抄録全体を表示
  • 古川 康一, 諏訪 正樹, 加藤 貴昭
    22 巻 (2007) 5 号 p. 563-573
    公開日: 2007/08/17
    ジャーナル フリー
    One of the main difficulties in motor skill acquisition is attributed to body control based on wrong mental models. This is true to various domains such as playing sports and playing musical instruments. In order to acquire adequate motor skill by modifying false belief, we need to help people find appropriate key points in achieving a body control and integrate them. In this paper, we investigate three approaches to realize such support. The first one is to encourage exploration of the relations among key points constituting a motor skill, using a technique of meta-cognitive verbalization. The second one is to represent a motor skill by appropriate mechanical models. The third one is to integrate rules for component tasks in achieving a compound task. These three approaches, we argue, help people build an integrated mental model consisting of multiple relations among various key points, one that seems to be indispensable for acquisition of motor skills. These ideas suggest the possibility to create new skill rules to perform difficult tasks automatically.
    抄録全体を表示
    • |<
    • <
    • 1
    • >
    • >|
feedback
Top