人工知能学会論文誌
Online ISSN : 1346-8030
Print ISSN : 1346-0714
検索
OR
閲覧
検索
22 巻 , 4 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
    • >
    • >|
論文
  • 原田 健, 佐久間 淳, 小野 功, 小林 重信
    22 巻 (2007) 4 号 p. 364-374
    公開日: 2007/04/13
    ジャーナル フリー
    Function optimization underlies many real-world problems and hence is an important research subject. Most of the existing optimization methods were developed to solve primarily unconstrained problems. Since real-world problems are often constrained, appropriate handling of constraints is necessary in order to use the optimization methods. In particular, the performances of some methods such as Genetic Algorithms (GA) can be substantially undermined by ineffective constraint handling. Despite much effort devoted to the studies of constraint-handling methods, it has been reported that each of them has certain limitations. Hence, further studies for designing more effective constraint-handling methods are needed.
    For this reason, we investigated the guidelines for a method to effectively handle constraints. The guidelines are that the method 1) takes the approach of repair operators, 2) monotonically decreases both the number of violated constraints and constraint violations, and 3) searches over the boundaries of violated constraints. Based on these guidelines, we designed a new constraint-handling method Pareto Descent Repair operator (PDR) in which ideas derived from multi-objective local search and gradient projection method are incorporated. Experiments comparing GA that use PDR and some of the existing constraint-handling methods confirmed the effectiveness of PDR.
    抄録全体を表示
  • 青山 一美, 南野 活樹, 下村 秀樹
    22 巻 (2007) 4 号 p. 375-388
    公開日: 2007/04/18
    ジャーナル フリー
    An autonomous agent in the real world should learn its own sensor-motor coordination through interactions with the environment; otherwise the behaviors can not be grounded and they can easily be inappropriate in the variety of the environment. The sensor-motor signals are usually complex time sequence, therefore the cognitive action system of the agent has to handle them.In this paper, we propose a computational model of the cognitive action system that consists of a sensor space HMM-SOM, a motor space HMM-SOM and connection mapping between the two HMM-SOMs. A HMM-SOM can be recognized as a set of HMMs that are placed in a SOM space. It can model a set of complex time series signal in a self-organizing manner.We apply this HMM-SOM based cognitive action system on vision-motion and auditory-articulation signals. The experimental results show that this system is basically capable of constructing sensor-motor coordination structure in a self-organizing manner, handling complex time series signals.
    抄録全体を表示
  • 前野 義晴, 大澤 幸生
    22 巻 (2007) 4 号 p. 389-396
    公開日: 2007/04/25
    ジャーナル フリー
    This paper addresses a new problem to infer an invisible fixer in an organization from communication (node discovery problem). Human-interactive annealing together with crystallization algorithm aims at inventing scenarios from the gap between prior understanding and observation. Four functions for ranking the relevance of the portion of observation, and two types of communication strength within an organization are studied. In the experiment, information relevant to identify an invisible fixer in an online decision-making environment is successfully retrieved.
    抄録全体を表示
  • 和泉 潔, 松井 宏樹, 松尾 豊
    22 巻 (2007) 4 号 p. 397-404
    公開日: 2007/05/02
    ジャーナル フリー
    We constructed an evaluation system of the self-impact in a financial market using an artificial market and text-mining technology. Economic trends were first extracted from text data circulating in the real world. Then, the trends were inputted into the market simulation. Our simulation revealed that an operation by intervention could reduce over 70% of rate fluctuation in 1995. By the simulation results, the system was able to help for its user to find the exchange policy which can stabilize the yen-dollar rate.
    抄録全体を表示
  • 染谷 博司, 山村 雅幸, 坂本 健作
    22 巻 (2007) 4 号 p. 405-415
    公開日: 2007/05/15
    ジャーナル フリー
    This paper discusses DNA-based stochastic optimizations under the constraint that the search starts from a given point in a search space. Generally speaking, a stochastic optimization method explores a search space and finds out the optimum or a sub-optimum after many cycles of trials and errors. This search process could be implemented efficiently by ``molecular computing'', which processes DNA molecules by the techniques of molecular biology to generate and evaluate a vast number of solution candidates at a time. We assume the exploration starting from a single point, and propose a method to embody DNA-based optimization under this constraint, because this method has a promising application in the research field of protein engineering.

    In this application, a string of nucleotide bases (a base sequence) encodes a protein possessing a specific activity, which could be given as a value of an objective function. Thus, a problem of obtaining a protein with the optimum or a sub-optimum about the desired activity corresponds to a combinatorial problem of obtaining a base sequence giving the optimum or a sub-optimum in the sequence space. Biologists usually modify a base sequence corresponding to a naturally occurring protein into another sequence giving a desired activity. In other words, they explore the space in the proximity of a natural protein as a start point.

    We first examined if the optimization methods that involve a single start point, such as simulated annealing, Gibbs sampler, and MH algorithms, can be implemented by DNA-based operations. Then, we proposed an application of genetic algorithm, and examined the performance of this application on a model fitness landscape by computer experiments. These experiments gave helpful guidelines in the embodiments of DNA-based stochastic optimization, including a better design of crossover operator.

    抄録全体を表示
  • 村木 雄二, 狩野 均
    22 巻 (2007) 4 号 p. 416-424
    公開日: 2007/05/15
    ジャーナル フリー
    In this paper, we propose a multiagent model for wide-area disaster-evacuation simulations with local factorsconsidered. Conventional multiagent models for evacuation simulations neither allow general-purpose computers to executewide-area simulations nor allow the object area to be changed easily. If these problems are solved, however, these simulations can be useful for local governments to make disaster damage prevention plan. In the proposed model, each roadis expressed by a series of cells. A computational amount relevant to interaction among agents is reduced by describing themodel for agents to be affected by other agents through the state of each cell. This makes possible wide area simulations. Using the data of a digital map database that is widely used for car navigation systems enables the simulations to beperformed for any region in Japan. Local factors are reflected in simulations by setting some parameters for evacuees, anevacuation environment, and disaster damage prevention plan of the object area. As an evaluation experiment, wesimulated the situations of Kobe city on the date of the Great Hanshin-Awaji Earthquake. Simulations results about thepercentage of evacuees who arrived at refuges were in good agreement with the actual data when parameters forevacuation-start timing were adjusted. We also simulated the current situations of two cities, Kobe and Tsukuba, and confirmed that this model was successfully applied to the two cities. From these evaluation experiments, we believe thatthis model can be applied to various areas and will perform further experiments in the future.
    抄録全体を表示
  • 徳久 良子, 寺嶌 立太
    22 巻 (2007) 4 号 p. 425-435
    公開日: 2007/05/24
    ジャーナル フリー
    In this paper, we investigate distinctive utterances in non-task-oriented conversational dialogue through the comparison between task-oriented dialogue and non-task-oriented conversational dialogue. We then found that Indirect Responses (IRs) and Clarification Requests (CRs) are significant in non-task-oriented conversational dialogue. IRs are cooperative responses to other's question, while CRs are clarification questions. We analyzed the rhetorical relations about IRs and CRs. We then found that the IRs are generated by evidence and causal relations, while the CRs are generated by elaboration relation and causal relations.
    抄録全体を表示
  • 楊 静宏, 五味 壮平, 川原 正広, 新貝 鉚蔵
    22 巻 (2007) 4 号 p. 436-443
    公開日: 2007/05/24
    ジャーナル フリー
    In this paper, we analyze the effect of the speed on the task of random number generation. Each subjectgenerates the sequences of ``random'' numbers under various speed conditions. In addition to the 16 indiceswhich have been used in the past studies, we introduce and calculate the new indices called the ``frequency ofschema''. From the statistical analysis, we find that there is an obvious change, along with the speed, in theproperties of sequences, indicating the transition of the underling information processing. It is shown that thefrequencies of schemas are effective indices for describing the properties of sequences produced under high-speed conditions.
    抄録全体を表示
  • 田口 亮, 木村 優志, 小玉 智志, 篠原 修二, 入部 百合絵, 桂田 浩一, 新田 恒雄
    22 巻 (2007) 4 号 p. 444-453
    公開日: 2007/05/24
    ジャーナル フリー
    Recently, studies on learning of word meanings by agents have begun. In these studies, a human shows objects to an agent and utters words such as ``red'' or ``box''. The agent finds out object's feature represented by each spoken word. In our method, firstly, the agent learns probability distribution p(x) and conditional probability distribution p(x|w), where x is an object feature and w is a word. If a word w does not represent a feature x, p(x) and p(x|w) will be almost same distribution because x is independent of w. This fact enables the agent to use distance between p(x) and p(x|w) when inferring which feature the word represents. Previous works also employ similar stochastic approaches to detect the feature. However, such approaches need a lot of examples to learn correct distributions.
    抄録全体を表示
  • 小笠原 昌子, 佐藤 眞木彦
    22 巻 (2007) 4 号 p. 454-460
    公開日: 2007/06/05
    ジャーナル フリー
    The mammalian immune system is a subject of great research interest because of its powerful information processing capabilities, namely, adaptivity. The adaptivity of the immune system is characterized by mainly two aspects, responsibility and diversity. The responsibility is a result of the response network of the immune system and the diversity is arise from gene rearrengement of the immune system. Recentry many artificial immune algorithms were devised by inspiring the adaptivity of the immune system. In terms of the two aspects of the immune system, however, those artificial immune algorithms only utilize the responsibility using models of response and regulation networks in the immune system. This paper proposes a new scheme of artificial immune algorithm, called Rearrangement Immune Algorithm (RIA), in which the rearrengement of the immune system is utilized combining evolution of the gelmline with an optimization of genetic algorithms. We empirically shows the effectiveness of our new scheme, RIA, with applying the Rosenbrock function and an HP folding problem.
    抄録全体を表示
    • |<
    • <
    • 1
    • >
    • >|
feedback
Top