An agent provides adaptability so that the agent can react changes of environments, select suitable behaviors for the environments, and execute the behaviors. To apply the adaptable agents to actual environments, systems to input and output to the actual environments are required. However, middlewares proposed in ubiquitous computing research area provide the system to input and output to the actual environments. In this paper, we introduce the middlewares, discuss how to apply the middlewares to the adaptable agents.
Computational Mechanism Design is a new field which combines social choice among agents who has private information and computer science concepts like computational costs and communication costs. We need knowledge for micro economics and game theory and knowledge for multi-agent systems and computer science. Further, Computational Mechanism Design is one of the possible studies which can be applied to real applications directy. In this paper, classic mechanism design concepts are reviewed. Then, the basic computational mechanism design problems like combinatorial auctions are shown. Finally, the state of art themes and issues in computational mechanism design are introduced.
Recently financial education has become more important because financial technology is highly developing and financial market is growing in importance. In this research, we apply Business Game method to financial education. Especially we focused on learning of asset allocation. As a result of intensive experiments, we found that (1) players learned not to take excessive risk through business gaming and (2) they recognize the importance of risk control by our experiments. These findings indicate that our approach is valid for financial education.
To some extent, we can easily know the topics of the world thourgh TV programs. On the other hand, it is not easy for many users to find the TV programs that offer unexpected interesting information. Then we constructed a “topic extraction agent” which extracts keywords that compose some topics from the texts of electronic program guides (EPGs) and shows the keywords to a user. We also implemented a TV-program navigation system by using this topic extraction agent. In this paper, we describe the architecture of this agent and consider the precision of keyword extraction and that of TV-program retrieval.
The Logic of Multiple-Valued Argumentation (LMA) is an argumentation framework that allows for argument-based reasoning about uncertain issues under uncertain knowledge. In this paper, we describe its applications to Social Web: SNS and Wikipedia. They are said to be the most influential social Web applications to the present and future information society. For SNS, we present an agent that judges the registration approval for mymixi in mixi in terms of LMA. For Wikipedia, we focus on the deletion problem of Wikipedia and present an agent that argues about the issue on whether contributed articles should be deleted or not, analyzing arguments proposed for deletion in terms of LMA.
In a keyword auction, advertisers submit their bids to search keywords and their ads are displayed according the result of the auction when people search the keyword on internet search engines. In existing keyword auctions, the number of slots is determined in advance and the obtained social surplus is not always maximized. Thus, we develop new keyword auction protocols in which the auctioneer can flexibly determine the optimal number of slots which maximizes social surplus. The first protocol is based on the Vickrey-Clarke-Groves (VCG) mechanism. Although this protocol has good theoretical characteristics such as strategy-proofness, determining the payment is quite complicated. The second protocol, which we call the GSP with Exculsive Right (GSP-ExR), eliminates these VCG drawbacks. If the value per click of the highest ranked bidder is large enough, then this bidder can exclusively display his advertisement by paying a premium.
CafeOBJ's verification method with proof scores which only uses reductions is presented. The example of QLOCK (mutual exclusion protocol with a waiting queue) is used to present theory and principle of the verification method, and techniques for writing the proof scores are also explained.
In ubiquitous computing scenarios, applications pro-actively support user activities by changing their behavior according to their contexts which for example represent the physical environment. Realizing such applications inherently calls for new development methodologies. In this paper, we present a survey on application development in ubiquitous computing and discuss future research directions in this field.
A cornerstone for delivering a proper user experience in ubiquitous/pervasive computing is the coordination of heterogeneous devices and services embedded into the environment. This paper presents a survey on such distributed coordination mechanisms, including techniques for service composition, resource/behavior adaptability, and for providing security/privacy. A particular focus has been put onto the analysis of what kind of issues and problems have been subject to research to meet the unique characteristics and demands imposed by ubiquitous applications. Based on the survey of existing work, this paper identifies issues that still remain to be tackled as well as directions of ongoing research activities.
This work deals with Q-learning in a multiagent environment. There are many multiagent Q-learning methods, and most of them aim to converge to a Nash equilibrium, which is not desirable in games like the Prisoner's Dilemma (PD). However, normal Q-learning agents that use a stochastic method in choosing actions to avoid local optima may bring mutual cooperation in PD. Although such mutual cooperation usually occurs singly, it can be maintained if the Q-function of cooperation becomes larger than that of defection after the cooperation. This work derives a theorem on how many times the cooperation is needed to make the Q-function of cooperation larger than that of defection. In addition, from the perspective of the author's previous works that discriminate utilities from rewards and use utilities for learning in PD, this work also derives a corollary on how much utility is necessary to make the Q-function larger by one-shot mutual cooperation.
A multi-agent system that consists of agents distributed in a network requires an application-level routing method using abstract identifiers such as agent names. We have accordingly proposed a routing method based on an Ordered-Tree-with-Tuft-shaped (OTT-shaped) overlay network. The method can maintain the network robustly and at low cost, and also make use of short-cuts between nodes to find shorter paths than those only found by the ordered tree of OTT. However it has a problem that the path-search cost by shortcuts is too expensive to achieve a reasonable success rate. This paper proposes our newly improved method that finds shorter-paths with both reasonable cost and high success rate of path-search by short-cuts and discusses some of experimental results achieved by software simulation. The results illustrate the validity of our proposed method.
Multi-issue negotiation protocols represent a promising field since most negotiation problems in the real world involve interdependent multiple issues. Our work focuses on negotiation with interdependent issue and, therefore, nonlinear (multi-optimum) agent utility functions. In this paper, we propose a new threshold adjustment mechanism in which agent who open their local information more than the others. In addition preliminary experimental results demonstrate that the threshold adjusting mechanism can reduce the computational cost, and the amount of private information that is required for an agreement among agent keeping enough optimality.
In general, ticket prices in theaters are fixed according to the quality of seats. However, it is difficult to set ticket prices properly reflecting consumers' reservation prices. Thus, the current fixed price does not attain the best efficient allocation. In this paper, we develop a model of seat reservation in theater by using auction theory. Then, we analyze differences between our model and fixed price model with multi-agent simulation. Additionally, we compare simulation results with Nash equilibrium that is derived under complete information. As a result, three important points are found: (1) the proposed model can make the number of sales larger than current one, (2) our model realizes higher social surplus than all fixed price model, and (3) comparing with fixed unique-price models where only unique price is set for several seats, the proposed model shows higher consumers' surplus and higher producer's surplus even in different situations.
Recently, automatic trading agents capture attention of traders. However, it is difficult to develop such agents for individual investors who have little programming skills. Thus, we proposed a support system for trading agents design for “Super kaburobo contest” which is a contest of automated stock trading. Consequently, the 60% of associate participants used the support system. By using the proposed system, the number of entries of traders who have little program skill increased. Additionally, it is confirmed that trading agents developed by the systems are comparable to Java-programed trading agents.
This paper develops two new false-name-proof auction mechanisms for hiring a team. In the problem of hiring a team, each agent is assumed to own one or more edges of a set system, and the auctioneer is trying to purchase a feasible solution to perform a task by conducting an auction. We introduce two models of false-name manipulations in hiring a team auctions and propose the MP and AP mechanisms, that are robust against false-name manipulations. Furthermore, we show the frugality ratio of MP is bounded by n2n, and that of AP is bounded by reserve cost, which is choosen a priori by the auctioneer.
In this paper, we propose enhanced approximation algorithms of combinatorial auction that are suitable for the purpose of periodical reallocation of items. Our algorithms are designed to effectively reuse the last solutions to speed up initial approximation performance. We present experimental results that show our proposed algorithms outperform existing algorithms in some aspects when the existing bids are not deleted. Also, we propose an enhanced algorithm that effectively avoids undesirable reuse of last solutions in the algorithm. This is especially effective when some existing bids are deleted from the last cycle.
Multiagent Partially Observable Markov Decision Process (Multiagent POMDP) is a popular approach for modeling multi-agent systems acting in uncertain domains. An existing approach (Search for Policies In Distributed EnviRonments, SPIDER) guarantees to obtain an optimal joint plan by exploiting agent interaction structure. Using SPIDER, we can obtain an optimal joint policy for large-scale problems if the interaction among agents is sparse. However, the size of a local policy is still too large to obtain a policy which length is more than 4. To overcome this problem, we extends the SPIDER so that agents can communicate their observation history and action history each other. After communication, agents can start from a new synchronized belief state thus the combinatorial explosion of local policies is avoided. Our experimental results show that we can obtain much longer policies as long as the interval between communications is small.
On-line planning agents can adapt to the dynamic environment by continuously modifying plans during the plan execution. However, it is not easy to sense the real world and obtain important information that might affect the current plan execution. This paper shows a new background sensing control method by which planning agents can effectively observe the real environment and obtain important information when necessary during the plan execution.
In this study, we investigate what would happen in a Chinese historical family line. We analyzed a particular family line, which had so many successful candidates, who passed the very tough examinations of Chinese government officials over 500 years long. First, we studied the genealogical records ‘Zokufu’ in China. Second, based on the study, we implemented an agent-based model with the family line network as an adjacency matrix, the personal profile data as an attribution matrix. Third, using “inverse simulation” technique, we optimized the agent-based model in order to fit the simulation profiles to the real profile data. From the intensive experiments, we have found that both grandfather and mother have a profound impact within a family to 1) transmit cultural capital to children, and 2) maintain the system of the norm system of the family. We conclude that advanced agent-based models are able to contribute to discover new knowledge in the fields of historical sciences.
An emergence of money is an important problem in economics. The economists have argued about this problem many years. However, the arguments were only abstract. In this research, we present a new model to explain a phenomenon that one of goods attains nature of money in the barter. Furthermore, we implement multi-agent simulator on the basis of this model and consider emergence of money in the complex social network by using the simulation result. Our new model which is different from a precedence research, consists of a micro-macro doubly structural network reflecting individual recognitions and social connections among agents. This model will show processes in which particular one of goods attains natures of money as a self-organization of the network. We examine this process by agent-based simulation with a static or dynamic social network and show that this model is able to apply to complex social network.