Automatic Inductive Programming Tutorial

Abstract

Computers that can program themselves is an old dream of Artificial Intelligence, but only nowadays there is some progress of remark. In relation to Machine Learning, a computer program is the most powerful structure that can be learned, pushing the final goal well beyond neural networks or decision trees. There are currently many separate areas, working independently, related to automatic programming, both deductive and inductive. The first goal of this tutorial is to give to the attendants a comprehensive view of the main areas related to the automatic induction of programs, a view which is not currently available to the community. ML researchers which do not know about Automatic Programming or researchers which work in just one of the areas would benefit from this tutorial. The expressivity of most Machine Learning languages (attribute-value) is basically equivalent to propositional logic, excluding work on ILP. The second goal of the tutorial is to show how we can go beyond these techniques by extending the expression power of the representation language. This can be done by adding elements programmers typically use, like variables, subroutines, loops, and recursion. This way, more complex problems can be addressed. The tutorial will start with a short overview of the different areas related to Automatic Programming. Most of the tutorial will focus on evolutionary / search-based techniques for generating programs.