Shin Ya Sato

Multi-Agent Systems Performance by Adaptive/Non-Adaptive Agent Selection

Our research interest lies in studying how local strategies about partner agent selection using reinforcement learning with variable exploitation-versus-exploration parameters influence the overall efficiency of multi-agent systems (MAS). An agent often has to select appropriate agents to assign tasks that are not locally executable. Unfortunately no agent in an open environment can understand the all states of all agents, so this selection must be done according to local information. In this paper we investigate how the overall performance of MAS is affected by their individual learning parameters for adaptive partner selections for collaboration. We show experimental results using simulation and discuss why the overall performance of MAS varies.

Adaptive agent selection in large-scale multi-agent systems

An agent in a multi-agent system (MAS) has to select appropriate agents to assign tasks. Unfortunately no agent in an open environment can identify the states of all agents, so this selection must be done according to local information about the other known agents; however this information is limited and may contain uncertainty. In this paper we investigate how overall performance of MAS is affected by learning parameters for adaptive strategies to select partner agent for collaboration. We show experimental results using simulation and discuss why overall performance of MAS varies.

Dependency of Network Structures in Agent Selection and Deployment

This paper shows that the statistical properties of the network topology are indispensable information for improving performance of multi-agent systems (MASs), though they have not received much attention in previous MAS research. In particular we focus on the applicability of the degree of an agent-the number of links among neighboring agents- to load-balancing for the agent selection and deployment problem. The proposed selection algorithm does not need global information about the network structure and only requires the degree of a server agent and the degrees of the nodes neighboring the server agent. Through simulation of several topologies reproduced by the theoretical network models, we show that the use of the local topological information significantly improves the fairness of the servers even for a large-scale network. We also find that the key mechanisms for load-balancing in a given network topology are highly asymmetric degree characteristics (scale- free) and the negative degree correlation.

Co-occurrence Analysis Focused on Blogger Communities

We studied the problem of finding a subspace of Web pages that is contextually consistent for co-occurrence analysis. We looked at blogs and proposed blogger-based co-occurrence analysis, which assumes that two items are relevant to each other if they appear in any of the blog entries posted by the same blogger. We show that (1) blogger-based analysis outperforms conventional page-based analysis in solving context-sensitive problems and that (2) analysis focused on bloggers forming a community yields better performance compared with that focused on isolated bloggers.

Generating extensional definitions of concepts from ostensive definitions by using web

We present GEO (Generating an Extensional definition from an Ostensive definition), a method to exhaustively gather items falling under an ostensively defined concept from the Web. By utilizing structural information about HTML documents, GEO automatically and efficiently gathers thousands of items from Web pages taking only 2 or 3 items as input. GEO also yields high precision (0.99 at maximum, 0.97 in average over a set of inputs). We also introduce a new style of searching information, called Item Search, in which GEO plays an essential role. Item Search can look for items in a targeted category that are the best matches against a given query. Some examples of Item Search are presented as the proof-of-concept of the idea.

An organization-related information maintenance component

This paper proposes an organizational information maintenance component (OIMC) that autonomously maintains organizational information (OI) in a multiagent system. Organization-related information is used to enable appropriate/satisfiable collaborations. It is strongly required in multi-agent (MA) applications that the agent, which is a delegate of some real-world entity or concept such as service, human, company and their relation (such as WARREN system whose agents represent customers, fund managers, technical analysts, and securities companies) has accurate and latest OI. There are also other MA systems that should take into account agents characteristics, such as honesty, reliability, strictness (for rating their task quality) and trustworthiness, for task allocations. The OI is, furthermore, likely to change over time because an agent may enter or exit a group, because data must adapt to human relations in the real world, and because data in their information servers becomes obsolete or is updated. Thus, agents must maintain their OI even while they are not active. However, since creating and maintaining organizational models of agents are not easy tasks, cooperative reasoners, planners, and schedulers with the mechanism for the OI maintenance are complicated and difficult to implement and maintain in actual applications. The proposed component, which works like a secretary and operates independent from the local agent and by exchanging OI, provides the general framework for maintaining an updated organizational model. The agents can ask organization-related queries to this component and can reflect the current OI to its inference.

Estimating Relevance of Items on Basis of Proximity of User Groups on Blogspace

We describe a new method to estimate the relevance of two items (such as products and works of art) on the basis of the relationship between the corresponding user (blogger) groups on a blogspace, where a user group refers to a collection of users interested in an item. We estimated the strength of the relationship between user groups on the basis of their proximity on the blogspace. We validated our approach through experimental studies using actual data. In developing the method for estimating relevance among items, we introduced a new technique for measuring the proximity of two groups of vertices on a network, which can be thought of as an extension of conventional co-occurrence analysis.