Yves Kodratoff

Knowledge Discovery in Texts: A Definition, and Applications

The first part of this paper will give a general view of Knowledge Discovery in Data (KDD) in order to insist on how much it differs from the fields it stems from, and in some cases, how much it opposes them.

A class of functions synthesized from a finite number of examples and a lisp program scheme

We define a class of functions that can be synthesized from example problems. The algorithmic representation of these functions is the interpretation of a given scheme. The instantiation of the scheme variables is realized by a new method which uses pattern matching then if necessary generalization and further pattern matching. One can compute the number of examples necessary to characterize in a unique way a function of this class.

Rating the interest of rules induced from data and within texts

Presents an application based on an evaluation of the interestingness of the rules induced from examples using inductive text mining (ITM). The better-known deductive text mining is called information extraction, and amounts to finding instances of a predefined pattern in a set of texts. ITM looks for unknown patterns or rules to discover inside a set of texts. We mainly discuss two of the problems of ITM: building ontologies of concepts, and extracting patterns.

Constructive Matching Methodology: Formally Creative or Intelligent Inductive Theorem Proving?

In this paper we explain why, and in what sense, the methodology for inductive theorem proving (IFTP) we develop is creative and we explain why our methodology cannot be said to be “intelligent”, as a human could be, and nevertheless it is suitable for a user-independent automatization of ITP.

Learning Expert Knowledge and Theorem Proving

After a brief historical recall, the paper describes, within the approaches issued from Artificial Intelligence, the different methodologies used in Machine Learning.

Logic programming and program synthesis

Well-known theoretical arguments proving that logic programming does not eliminate the problem of transforming a specification into an executable program are presented and illustrated. PS (program synthesis) is therefore still a real problem to be faced by AI (artificial intelligence) research since complete automation of a PS tool is still far off, especially for long, complicated specifications such as those usually met in practice. It is shown how it is quite possible to write down specifications in Prolog. Nevertheless, it is pointed out that well-known theoretical reasons limit this possibility, and a detailed analysis of the practical reasons why a formal specification may be hard to program in Prolog is provided. The present work contributes to the clarification of the exact role of PS in AI and in software engineering and its possible application to software certification.<<ETX>>