Abstract. The article deals with the problem of learning incrementally ('on-line') in domains where the target concepts are context-dependent, so that changes in context can produce more or less radical changes in the associated concepts. In particular, we concentrate on a class of learning tasks where the domain provides explicit clues as to the current context (e.g., attributes with characteristic values). A general two-level learning model is presented that effectively adjusts to changing contexts by trying to detect (via 'meta-learning') contextual clues and using this information to focus the learning process. Context learning and detection occur during regular on-line learning, without separate training phases for context recognition. Two operational systems based on this model are presented that differ in the underlying learning algorithm and in the way they use contextual information: MetaL(B) combines meta-learning with a Bayesian classifier, while MetaL(IB) is based on an instance-based learning algorithm. Experiments with synthetic domains as well as a number of 'real-world' problems show that the algorithms are robust in a variety of dimensions, and that meta-learning can produce substantial increases in accuracy over simple object-level learning in situations with changing contexts.Keywords: Meta-learning, on-line learning, context dependence, concept drift, transfer
MotivationThe fact that concepts in the real world are not eternally fixed entities or structures, but can have a different appearance or definition or meaning in different contexts has only gradually been recognized as a relevant problem in concept learning. Michalski (1987) was one of the first to formulate it; he suggested a specialized two-tiered representation formalism to represent different aspects of context-dependent concepts (see also Bergadano et al., 1992). Recently, context dependence has been recognized as a problem in a number of practical machine learning projects (e.g., Katz et al., 1990;Turney, 1993;Turney & Halasz, 1993;Watrous, 1993;Watrous & Towell, 1995; see also Kubat & Widmer, 1996). There, various techniques for context handling were developed. Most of these methods either assume that contextual attributes are explicitly identified by the user, or require separate pre-training phases on special data sets that are cleanly separated according to context.We are studying the effects of context dependence and changing contexts in the framework of incremental (or on-line) learning, and we are interested in learners that can adapt to different contexts without explicit help from a teacher. The scenario is as follows: assume that a learner is learning on-line from a stream of incoming (labeled) examples. Assume further that the concepts of interest depend on some (maybe hidden) context, and that changes in this context can induce corresponding changes in the target concepts. As a simple example, consider weather prediction rules, which may vary drastically with the change of seasons. The visible effects of such changes ...