Incremental Evolution in Genetic Programming

This paper presents a study of different methods of using incremental evolution with genetic programming. Incremental evolution begins with a population already trained for a simpler but related task. No other systematic study of this method seems to be available. Experimental evidence shows the technique provides a dependable means of speeding up the solution of complex problems with genetic programming. A novel approach that protects against poor choices of problem simplifications is proposed, improving performance. Testing performed on tracking problems of multiple stages is analyzed. 1. Introduction Some researchers in the field of computer learning have expressed skepticism that genetic programming (GP) will scale up from toy problems. This concern comes in part from the difficulty of solving problems with chicken and egg dilemmas. A chicken and egg dilemma occurs when the interaction of multiple behaviors aids in the performance of the task despite the fact that none of the beha..