Knowledge representation and organization in machine learning. (Q1210889)

The articles of this volume will not be indexed individually. The book contains 15 detailed papers from the Workshop on Knowledge Representation and Organization in Machine Learning held in 1987 in Germany. The first paper by W. Swartout and S. Smoliar shows how the requirement for a maintainable knowledge base can be used as a guideline for an expert system's representation. The Explainable Expert System offering a KL-ONE-like representation formalism for building a representation language is described. The second paper by W. van de Velde presents a two-level expert system: a representation for causal relations of a domain can be transformed into another one which is better suited for quick problem solving. The third paper by M. Mohnhaupt and B. Neumann discusses the requirement of learnable representations exemplified by a vision system. The transformations are from the original close-to-pixel representation to a prototypical representation of car movements. The fourth paper by R. Rada and H. Mili deals with a reorganization of different databases with different indexing schemata into one database. The procedure is based on occurrences of the same term in both databases and uses analogy for mapping two indices. In te fifth paper D. Litman observes an expert system designer at his work and proposes some categories for cognitive entities and operations. The sixth paper by K. Morik presents a unifying view of knowledge acquisition and machine learning. The reversibility of knowledge is stressed and conclusions for an integrated system are drawn. The next paper by Y. Kondratoff and G. Tecuci presents a system which integrates knowledge acquisition by questioning the user, machine learning using various techniques, and a performance element. The following paper by W. Emde examines requirements for knowledge representation formalisms posed by different learning tasks and demonstrates an inference machine specially designed to support incremental machine learning and knowledge revision. S. Thieme in the next paper discusses the knowledge which a machine learning algorithm needs. The tenth paper by M. van Someren is devoted to the dependency of the representation of examples from which the system learns and the learning results. The next paper by M. Manago and J. Blythe deals with the aspect of representation of the learning results by showing the implications of enhancing the representation of the concepts to be learned. Learning by analogy is investigated in the following paper by C. Vrain and Y. Kondratoff with respect to problem solving. Similarity-based learning by analogy is presented which makes good use of the dissimilarities of analogous problems. The thirteenth paper by D. Wilkins stresses the value of learning for knowledge base repair. In the next paper M. Pazzani combines similarity-based learning and explanation-based learning and shows that the representation needed for explanation-based learning can be built up by similarity-based learning. The last paper by S. Wrobel provides the model-driven rule-learning. It deals with both knowledge-based repair and the introduction of new cocepts into representation language of the domain theory. The book may be regarded as very useful for various classes of researches in AI, and the reader may feel inspired to think some more about the impact of representations on machine learning.