Explain GP in Introduction

[fergunet]

You can adapt this text I wrote in the StarCraft paper (following the structure and adapting it to BYOD topic, and adding references to classification instead behaviour modelling):

GP belongs to the category of \emph{Evolutionary Algorithms} \cite{EAs_Back96}, optimization techniques inspired by natural evolution. This method is commonly used to create solutions internally encoded as trees or linear graphs \cite{Brameier2007}, but it has also been used for the generation of Turing-complete programs. These algorithms are normally used to generate computer programs to perform an objective task, optimizing a metric or objective functions called {\em fitness}. This technique can produce combinations of conditions and actions that are potentially very different from what a human programmer could design, making it possible to obtain competitive bots from scratch, i.e. without adding human knowledge. This introduces a high difference with respect to the usual improvement of behavioral models by mean of EAs \cite{offline-evolutionary-learning,colomiCEC2004,Genebot_CEC11}, which is based on the optimization of parameters that guide the bots behavior, and consequently constrained to a human-designed model and its possible limitations.

[unintendedbear]

Perfect, thank you!