Combined Learning Methods and Mining Complex Data

RSCTC 2010 special session

Motivation and Goals:

Data mining and machine learning have shown tremendous progress in the last decades. Numerous methods have been introduced to discover different representations of knowledge from data and the number of their applications in various fields is growing. Nevertheless, many of approaches are based on using a single learning algorithm to static and standard forms of data (mainly tabular forms). However, the rapid growth of information technology gives access to more complex, larger and, generally speaking, “more difficult” data sets that pose new challenges for researchers and ask for a variety of dedicated approaches. Hence, in this session we would like to focus on data characteristics related to the following modern challenges:

Processing data streams and learning in changing and distributed environments. Many sources generate continuous and time changing data, where the data distributions and target concepts change over time. Mining them and adapting to concept drifts is of great interest.
Semi-supervised learning. In many domains only a limited number of labelled examples is available, therefore learning approaches should take as much advantages of unlabelled data as possible to produce the efficient solution. In case of classification, it leads to development of such approaches as co-training or active learning. On the other hand, supervised information is also taken into account in new clustering algorithms.
Learning from imbalanced data, where one class contains much smaller number of examples than the remaining classes. The imbalanced distribution of classes constitutes a difficulty for standard learning algorithms and calls for specialized approaches.

The other aim of this session is to promote complex learning methods such as combining several strategies, with classifier ensembles being of particular interest. Although the multiple classifiers are still “young” they have already proved to be accurate, flexible and sometime more efficient than single classifiers. The scope of this session also covers other combined methods such as regression ensembles, hierarchical classifiers or meta-learning, and their integration with feature processing.

Both methodological and application oriented papers fit the session. In particular, we encourage researchers to study the use of combined learning methods for above mentioned challenges. Other applications are also encouraged, including relationships to generalisations of rough sets for knowledge discovery.

The main aim of this session is to gather researchers interested in these issues and to demonstrate results in these and related areas of mining difficult data.

Topics of interest:

Supervised ensemble learning problems and multiple classifier systems
Theoretical aspects of constructing combined learning systems
Combined learning methods for regression and ordinal data
Classification, clustering and frequent patterns from data streams
Ensemble learning in changing environments
Detecting changes and concept drift in evolving data
Knowledge discovery from ubiquitous environments
Incremental online learning algorithms
Active learning and co-training
Semi-supervised approaches (including clustering)
Sampling techniques for imbalanced data
Handling class imbalance by modifying inductive bias and post-processing of learned models
Pre-processing, structuring and organizing complex data
Applications, especially in data mining, medicine, text processing, web mining, image or multimedia analysis, bioinformatics.

Organisers, contact:

Jerzy Stefanowski - Jerzy.Stefanowski@cs.put.poznan.pl
Poznań University of Technology, Institute of Computing Science
ul. Piotrowo 2, 60-965 Poznań, Poland
phone: +48 61 6652933
www.cs.put.poznan.pl/jstefanowski

Program Committee of the Session:

Włodzisław Duch Nicolaus Copernicus University, Toruń, Poland	Krzysztof Krawiec Poznań University of Technology, Poland
Joao Gama Univeristy of Porto, Portugal	Ludmila I. Kuncheva University of Wales, Bangor, UK
Jerzy Grzymała-Busse University of Kansas, Lawrence, USA	Stan Matwin University of Ottawa, Canada
Nathalie Japkowicz University of Ottawa, Canada	Ernestina Menasalvas Technical University of Madrid, Spain
Jacek Koronacki IPI PAN, Warsaw, Poland	Zbigniew Raś University of North Carolina – Charlotte, USA
Wojciech Kotłowski Centrum Wiskunde & Informatica, Amsterdam, Netherlands	Alexey Tsymbal Corporate Technology Division, Siemens AG, Erlangen, Germany