Data Aggregation and Multi-Scale Analysis of Complex Systems
Speaker: Professor Yves Demazeau, President of the French AI Association (AFIA), Director of Research at CNRS (Centre national de la recherche scientifique).
Date and Time: 14th January, 2015, 1:30 – 2:30pm
Seminar Room: CB25.GD.02
Seminar Chairman: Associate Professor Jinyan Li, AAI (Jinyan.li@uts.edu.au)
Abstract
The analysis of complex systems often requires a multi-scale approach. The events necessary for the explanation of the behavior of such systems can receive different interpretations depending on the granularity of information available for the analysis. We are interested in data aggregation methods to provide relevant information to detect crucial events occurring at different spatial and temporal scales.
The problem of multi-scale representations requires to address two fundamental aspects of aggregation: its information properties and its structural properties. We show that aggregation based on the optimization of two opposite properties, to minimize the complexity of the dataset while preserving as much as it can its explanatory power. The information theory provides the mathematical basis for the formalization of this optimization problem. But the aggregation must also preserve the essential properties of the system studied; these properties can be expressed as algebraic constraints structuring the search space. We finally present a generic algorithmic framework for resolving the different versions of the problem.
Two application frameworks serve to illustrate the proposed method: the aggregation of geographic data for the analysis of international relations and the aggregation of execution traces to detect anomalies in distributed systems.
Biography of speaker: Yves Demazeau received his PhD in Computer Science at Grenoble National Polytechnic Institute in 1986 and his HDR at the same university in 2001. He is currently Director of Research at CNRS and External Professor at the Faculty of Science and Engineering of the University of Southern Denmark since 2002. He has been Visiting Professor at VUB Brussels in 1989, Southern Denmark University in 1994, and University of Technology, Sydney in 2014. Yves Demazeau has been active in the research and development of Multi-Agent Systems (MAS) for 27 years. He has edited 27 books or proceedings, has authored or co-authored 151 papers and has given more than 46 invited lectures, 141 seminars, and 22 tutorials in this research area. With a close connection to his research interests, Yves Demazeau has advised 51 MSc and 28 PhD theses. He has also participated to 78 PhD examination boards and 13 HDR examination boards. From design to development, his main research contributions in the MAS have been the introduction of interaction protocols for Agent Communication Languages as adopted by the FIPA, the introduction of the concept of organisations as a ground component of MAS, and the overall VOWELS multi-agent oriented framework. He is also well known for the large-scale and original applications he has developed together with his research team in various domains: autonomous robotics, computer vision, language processing, geographical information systems, telecommunications, interactive games, artistic creation and web intelligence. Since 2004, Yves Demazeau has expanded his research to more general Artificial Intelligence issues. His current basic research activities in Artificial Intelligence address four themes: user modelling, privacy and trust, MAS dynamics, and MAS evaluation. Applications are also at number of four: smart networks, service to the person, production management, and bio-robotics. Yves Demazeau has been teaching MAS in Western Europe (AGH Krakow, UJF Grenoble, UL Luxembourg, UPD Paris, UPM Madrid, USD Odense), and in Southern America (UFC at Fortaleza, UFRGS Porto Alegre, UNICEN Tandil) for more than 25 years. He has served or is serving on the program committees of 240 conferences and workshops. He was the co-originator in 1989 of the MAAMAW workshop series and in 1999 of the CEEMAS workshop series. He was cofounder of the ICMAS conference series, that has been later transformed into the AAMAS conference series. He has been member of the Board of the International Foundation on Multi-Agent Systems (IFAAMAS) from 1998 until 2002, and the co-originator and research coordinator from 1998 until 2002 of AgentLink, the CEC European Network of Excellence in the area of (multi-) agent-based computing. He currently holds the chairing of the PAAMS series since 2009. Yves Demazeau has served or is serving as editorial board member for the Journal of Autonomous Agents and Multi-agent Systems (JAAMAS, Springer Verlag), Multi-Agent and Grid Systems (MAGS, IOS Press), Revue d'Intelligence Artificielle (RIA, Hermes, currently serving as Editor-in-Chief), and of the book series on Multi-Agent Systems, Artificial Societies and Simulated Organisations (MASA, Springer Verlag). He is President of the French AI Association (AFIA) since 2011.