ADMIT: Agent-oriented Distributed Data Mining using Computational Statistics
I worked as a Postdoctoral research fellow in this FP7 Marie-Curie Actions Intra-European Fellowships (IEF) Project.
ADMIT Project is executed since 1-st June 2011 - 18th July 2014 at Clausthal University of Technology, at the Department of Informatics, at the Chair for Business Information Technology in the Mobile and Enterprise Computing research group and supervised by the director of the Department of Informatics and the head of the Mobile and Enterprise Computing research group Prof. Jörg P. Müller.
Today's systems for managing critical infrastructure such as traffic, energy, or industry automation systems are highly complex, distributed, and increasingly decentralized. Multi-agent systems (MAS) provide an intuitive metaphor and configurable, robust and scalable methods for problem-solving and control in distributed, decentrally organized system. The purpose of Distributed Data Mining (DDM) is to provide algorithmic solutions for data analysis in a distributed manner to detect hidden patterns in data and extract knowledge necessary for decentralized decision making. A new promising area of research studies possibilities for coupling MAS and DDM by exploiting DDM methods for improving agents’ intelligence and MAS systems performance. In the ADMIT project we focus on methods for distributed estimation of parameters for the individual agents, agent groups, and system-level information models. Our approach is based on Computational statistics (CST), which includes a set of methods for approximate solution of statistical problems without complex statistical procedures. The goal of the ADMIT project is to develop an agent-oriented DDM framework, which includes a set of computationally effective, robust and easy to apply methods for models parameter estimation and allows easy incorporation into MAS applications to analyze models at different levels of MAS.
The scientific research objectives
Objective 1: To develop a conceptual architecture of agent-oriented DDM framework as well as a methodology of its usage in multi-agent programming frameworks;
Objective 2: To develop a set of computationally effective and reliable to bad data quality CST-based DDM methods, for efficient estimation of model parameters on the basis of distributed data as well as estimate the methods performance;
Objective 3: To assess the impact of incorporation of the DDM framework to MAS-based applications (with the main focus on traffic and logistics domains).
Summary of Progress and Details for each Task
Objective1 (Tasks 1-2): To develop a conceptual architecture of an agent-oriented (AO) distributed data mining (DDM) framework and a methodology of usage in the MAS frameworks, the necessary data flows and use cases were described and analysed. Considering traffic routing problems, several scenarios were described involving tactical and strategic planning of traffic participants’ behaviour [14, 15]. The necessary data flows for the comparison of routes [1, 8, 12], change-point analysis of traffic state [5, 11], and travel time forecasting [6, 9, 10] and clustering [2, 3, 4] were investigated. The necessary AO models for DDM layer were developed that insure the intelligent behaviour of agents and their collaboration. The conceptual decentralized, distributed and centralized architectures of the AO DDM framework are described in [1-10, 14, 15]. Finally, such Future Internet capability as cloud-computing architecture was considered for decentralized traffic management system, including the concrete scenarios and distributed data flow processing and mining methods [1, 4, 7,].
Objective 2 (Tasks 3-4): A set of computational statistics (CST) based methods for the efficient estimation of model parameters on the basis of distributed data sources was developed, which are computationally effective and reliable to bad data quality. The performance of the methods was estimated. In [14, 15] CST-based methods were applied to decision making strategies. In [1, 8, 12] a resampling based approach to the comparison of routes in a graph was suggested and its efficiency was estimated. In  a change-point problem that supports agents in detecting changes in their environment was described and CST-based solution methods were developed. The performance of these methods was analyzed in . CST-based methods were developed to support agent cooperation for forecasting with regression model [4, 6, 9, 10] and for clustering [2,3] of travel times by intelligent agents;
Objective 3 (Tasks 5-6): The impact of incorporating the DDM framework into MAS-based applications in traffic and logistics domains was assessed. All the scenarios [1-10, 14, 15] were taken from the domains mentioned. All methods were validated using experiments using data from real traffic networks. These initial data were obtained from the on-going project PLANETS of NTH, where our research group at TU Clausthal participates. This collaboration with other MEC Group colleagues implied the co-authored publication of Dr. Fiosina.
The results of these objectives were presented by Dr. Fiosina and co-authors at 11 international conferences. 16 project-related papers were published during the reported period (See the next section).