Machine learning: ECML '97. 9th European conference on Machine Learning, Prague, Czech Republic, April 23--25, 1997. Proceedings (Q1356164)

These are the proceedings of the ninth European Conference on Machine Learning (ECML-97), held in Prague, Czech Republic, in April 1997. The volume contains 13 long papers and 13 short papers on virtually all aspects of machine learning, as well as full versions of two invited papers. The following topics are dealt with: concept learning of autonomous agents on the basis of language games, the relation between the compression of the data sample and the prediction performance based on the notion of Kolmogorov complexity, the induction of feature terms, the use of qualitative knowledge in a so-called soft teaching approach for the improvement of skill acquisition, the augmentation of reinforcement learning by planning, a combination of Plotkin's notion of ``(relative) least general generalization'' with Vere's structural matching method, a new classification algorithm based on voting feature intervals, the construction of intermediate concepts by decomposition of real functions, a consideration of conditions for Occam's Razor applicability and noise elimination, combining the neural network and iterative attribute construction in order to learn relatively short but complex Boolean combinations and prototypical structures, a discussion of metrics for concept clustering, an algorithm that learns when negative examples heavily outnumber the positive examples, a model for generalization based on confirmatory induction, learning linear constraints in inductive logic programming, a finite-element method for continuous reinforcement learning problems, inductive genetic programming with decision trees, parallel and distributed search for mining multivariate time series of categorial data, compression-based pruning of decision lists for propositional domains, probabilistic incremental program evolution based on a stochastic search through program space, extraction of oblique decision rules from neural networks, a knowledge discovery system that induces and uses decision rules in the context of large relational databases, optimality considerations for Markovian decision problems, model combination in a multiple-data-batches scenario, a search-based methodology for class discretization, the use of natural ideal operators in inductive logic programming, a case study in loyalty and satisfaction research, a reinforcement learning approach to (robot) group learning, and global data analysis and the fragmentation problem in decision tree induction. Also included are four workshop position papers on case-based learning, empirical learning in a natural language processing framework, human-agent interaction and machine learning, and learning in dynamically changing domains.