DM 2014 - Special session: Data mining with meta-learning and hierarchical architectures

Special session: Data mining with meta-learning and hierarchical architectures
The IEEE World Congress on Computational Intelligence, Beijing, China, 6-11 July 2014
Goals of this special session:
This session will be devoted to approaches that in an intelligent way integrate various components of learning algorithms used for data mining, especially in meta-learning, multimodel architectures, facilitating knowledge transfer and deep learning. Integration of machine learning algorithms becomes increasingly more important, especially in applications to hard problems which still wait to be solved, where application of specialized methods that do not use additional knowledge has led to limited success. Hard problems and big data need much more than single neural network or single learning machine. Sophisticated data transformations play more and more important role. Data mining packages contain hundreds of algorithms that may be composed in millions of ways. Automatization of this process requires analysis of learning algorithms at the meta-level. Methods that extract various forms of useful knowledge, share and integrate it for intelligent information processing, are necessary to solve hard problems. Such methods may be inspired by the organization of the brain, or may be based on formal algorithms. One promising direction is to use methods that construct new features, learning from successes of different algorithms, extracting knowledge from indirect, partial learning and using it to build final potential solutions. Another interesting aspect in the construction of complex computational intelligence methods is dealing with different levels of abstraction; useful meta-knowledge may come in the form of highly abstract heuristic knowledge directing the search process for the optimal model, or may be hidden in the details of algorithm implementation.
The main subjects of interest are:
? Meta-learning algorithms and system architectures.
? Meta-knowledge representation, acquisition, application, re-use and construction, analysis of the usefulness of knowledge.
? Knowledge transfer, knowledge sharing, transfer learning.
? Meta-learning for big data.
? Multimodel architectures, integration of hierarchy of individual models for data mining.
? Multimodel data mining systems/algorithms, which integrate several methods of data analysis at different levels of granularity.
? Data mining that use hybrid/heterogenous models
? Advanced architectures of data mining systems. Combinations of machine learning, neural networks, fuzzy systems, etc.
? Transformation-base learning, including deep learning algorithms.
? Extraction and construction of new features that simplify the complex learning process, including pre-processing methods, multimodal signal processing, extraction of information from specific types of data.
? Methods of reasoning for automatic creation of decision models, estimation of usefulness of knowledge for a given problem.
? Applications to challenging problems, methods for testing complex systems.
The session is not strictly limited to the above subjects. Every aspects of meta-learning or other integration of learning algorithms and knowledge are welcome.
IMPORTANT DATES
December 20, 2013:
Paper submission deadline.
March 15, 2014:
Notification of paper acceptance.
April 15, 2014:
Final manuscript submission deadline.
Organizers:
Norbert Jankowski and Wlodzislaw Duch
Department of Informatics, Nicolaus Copernicus University
ul. Grudziadzka 5, 87-100 Toruń, Poland
norbert-AT-is.umk.pl or
wduch-AT-is.umk.pl