Learning Geometric Concepts with an Evolutionary Algorithm.

Abstract

We present a system that is able to learn descriptions of pictures with an evolutionary algorithm approach. The descriptions are programs in a turtle-graphics language and the described pictures are scenes from an environment with a robot-arm acting in a blocks-world. A measure of similarity of pictures is presented which can be computed fast and supplies gradient information with regard to translation, rotation and expansion of objects. The algorithm is very qualified for the classification of large sets of pictures as objects which, once recognized, are re-found quickly. Even if they are in different shapes and orientations or in large composed scenes. The approach differs from genetic programming as three simple problem specific operators are used instead of crossover. 1 Introduction An important property of intelligence is the ability to learn. Therefore, it is interesting to build and investigate artificial learning systems. At the Universitat des Saarlandes systems are develope...