Machine learning - EWSL '91. European working session on learning, Porto, Portugal, March 6-8, 1991. Proceedings (Q1202143)

The articles of this volume will not be indexed individually. From the editor's foreword: ``During the last decade numerous learning techniques, such as empirical induction, explanation-based learning, version spaces, etc., have been developed and tested on a variety of applications. It appears that each method leads to interesting results, but that real applications require a combination of the various methods in order to solve practical problems. This is why Part 1 of this book is concerned with a topic which has been emerging during the very last few years in Machine Learning (ML), multistrategy learning, also called constructive learning. This part of the book starts with a kind of discussion between a theoretical approach as given by Giordana et al., and Tecuci's more intuitive approach. Part 1 goes on with several intances of a multi-strategic approach to ML, and ends up with a discussion on representation changes steered by F. Bergadano. The topic called discovery can be said to be an application of multi-strategy learning, since it tends to use a combination of inductive and deductive techniques in order to discover functional relationships among sets of number data. This topic has been already extensively dealt with in statistics. The point of view of ML in that matter, as shown by papers of Part 2, brings to the fore the importance of using the symbolic information implicitly contained in the data, rather than the raw data alone. Part 3 deals with the numeric and statistical approaches, including techniques using information compression as the key to learning, also often called inductive construction of decision trees by ML specialists. The last paper in this part is the only representative of genetic algorithms that passed the criticism of our references. Part 4 brings us back to purely symbolic techniques stemming from theorem proving techniques, including explanation-based learning, which have become so important since the middle of the last decade. Part 5 deals with inversion of resolution, which is of course an inductive technique, but the origin of which is so much embedded within logic programming that it counts as the very first inductive technique to be as well formalized as the deductive ones. In Part 6, analogy and case-based reasoning, we come back to more intuitive learning methods that have been undergoing a continuous progress during this last decade. Part 7, multiagents, contains two papers describing how agents may interact in order to improve their learning. This last topic has received very little attention up to now. Together with multi-strategy approaches they may well become the great research topics of the 1900s. Last part, applications, describes some applications to ML''.