Publisher's Synopsis
The principal message of this work is that the more forms of regularity a system is equipped to recognise and use, the greater its ability to draw conclusions and make predictions from its experience. This simple insight is given formal expression in an algorithm for generating and evaluating the possible logical forms of regularity.;Five types of knowledge are identified that allow sound analogical inference and hence good analogical performance, thus solving the traditional logical and computational problems of analogy. Programs are described for acquiring and using this knowledge in computer systems, and the approach is shown to be considerably more efficient than rule-based reasoning for some important tasks. Its extension to the general problem of induction generator, formally an inductive method that encompasses Goodman's revolutionary theory, promises to greatly extend the power of machine learning systems.
