CFP

Machine Learning Journal
Special Issue on
Meta-Learning

C. Giraud-Carrier, R. Vilalta and P. Brazdil, Guest Editors

CALL FOR PAPERS

MOTIVATION AND RESEARCH ISSUES

The proliferation of algorithms in Machine Learning (ML) and the growing
interest in Data Mining (DM) have created a need for techniques and tools

that facilitate the use of ML by novice users (e.g., to select adequate
algorithms for specific business problems). Such tools will facilitate
the transition of ML from research labs into industry.

Discovering new algorithms (or versions thereof) has occupied much of the

research of the past decade with reasonable success. Despite empirical
studies comparing various algorithms, however, much remains to be learned
about what makes a particular algorithm work well (or not) in a particular
domain. There is a need to formulate or acquire such meta-knowledge, and
make consistent use of it.

Although the term meta-learning has been ascribed different meanings by
different researchers, for the purpose of this special issue, meta-learning
is defined as any attempt to learn from the learning process itself. The
goal is to understand how learning itself can become flexible and/or
adaptable, taking into account the domain or task under study.

Papers are solicited on the following subjects:

• EXPLOITING META-LEARNING
  . Selection of ML/DM algorithms from a given set
  . Selection of parameterized versions of a particular ML/DM algorithm
  . Selection of pre-/post-processing tasks for a given set of ML/DM
  algorithms
  . Flexible/adaptable combination of ML/DM algorithms (e.g.,
  combinations of classifiers)
  . Flexible/adaptable design of complex system from basic parts (e.g.,
  combination of pre-processing tasks and DM steps)
  . Automatic shift of bias

• FOUNDATION OF META-LEARNING
  . Evaluation and comparison of meta-learners, including multi-
  criteria (i.e., beyond predictive accuracy)
  . Theoretical studies of algorithms' performance (including studies
  focused on particular components of algorithms) and their impact
  on meta-learning
  . Empirical studies of algorithms' performance (including studies
  focused on particular components of algorithms) and their
  contribution to meta-learning
  . Methods for dealing with small amounts of meta-data (including
  planning systematic experiments and ways to reduce meta-data
  acquisition costs) and their utility for meta-learning

• ISSUES WITH META-DATA
  . Crafting domain/task characteristics and their relation to
  learning performance
  . Theoretical and/or empirical studies reporting on domain/task
  characteristics that are relevant and not to the meta-learning
  process

The SUBMISSION DEADLINE is JUNE 8, 2002.