Home
Purpose of Conference
Topics
CISIM Dates
PROGRAMME
Keynote Speakers
Proceedings
Author Instructions
Paper Presentation
Paper Submission
Committees
Venue
Fees
Access
Accommodation
Postconference Publication
Previous Conferences

Biography:

Anna M. Bartkowiak has received her MSc and PhD degrees in Applied Mathematics from the University of Wroclaw, Poland, and her DSc degree (habilitation) from the Institute of Computer Science, Polish Academy of Sciences, Warsaw, Poland. She has worked several years as researcher in the Polish Academy of Sciences, Mathematical Laboratory in Wroclaw. Next she became academic teacher at the University of Wroclaw, firstly in the Mathematical Institute, and next in the Institute of Computer Science, where she got the post of Professor in computational statistics. She helt for 3 years the post of vice-dean of the Faculty of Mathematics and Informatics at the Wrocław University. She is presently affiliated as Professor Emeritus in the Institute of Computer Science, University of Wroclaw, Poland. She is a Fellow of the Royal Statistical Society, London, also a member of IBS, ISCB, IASC ISI, and of ENNS. She had been an active member of the International Biometric Society (IBS): she has served in the past for 6 years as Scientific Council member, for 6 years as the Award and Grant Committee member, and for 6 years as Conference Advisory Committe member. She has been also the founder and for the first 9 years the Chair of the Polish National Group of the IBS. Her scientific interests are: algorithms of computational statistics and multivariate analysis, in particular: graphical visualization of observational data, pattern recognition and neural networks. She has published, in Polish, several books on these topics. She has also published, in English, in international journals and conference proceedings - more than 200 papers on the mentioned topics of her interests.

Keynote Title:

Anomaly, Novelty, One-Class Classification

Abstract:

In data analysis and decision making we are frequently put int the position to judge whether the observed data items are normal or abnormal. This happens in banking, fraud credit card use, diagnosing a patient´s health state, fault detection in an engine or device like an off-shore oil platform or a gearbox in an airplane engine. We have to know when the system´s behavior may be judged as normal, and when is starts to be abnormal. Sometimes the "normal cases" are boring and only the "abnormal cases" are of interest. Usually it happens that the 'normal state' has a good representation, while the abnormal cases are rare and the abnormal class is ill-defined, thus we have to judge on abnormality using information from the normal class only, which is referred to as 'one-class classification' (OCC). In the paper I will give a survey of some recent research on the problem of dealing the OCC problem. In the last decade it became obvious that one-class classificators are needed in many important domains of our life environment. Such situations happen in everyday practice: an unexpected eruption of a volcano, a leakage of an off-shore platform in the Mexican Gulf, an outbreak of an atypical flue decimating the population, a plane crash. We need monitoring systems using non-invasive measurements able to signalize that something abnormal starts to happen. There is also the need of designing autonomous robots working in an vision-based environment, able to detect novelty and concentrate to explore further this novelty. What is the preferable method for designing the one-class-classifier and finding the unusual? It depends on the data. There is no single preferable method. One is sure: OCC is essential for our XXI century reality. I will show also an example of real data analysis: how to detect a masquerader (non-legitimate user) in a computer system -- when considering sequence of commands several thousands long. This will be shown considering Schonlau data.