Background

Contents

To give the reader a perspective on what characterises LCS exactly, and to which level they are theoretically understood, this chapter gives some background on the initial ideas behind designing LCS, and describes what can be learned from their development over the years and the existing theoretical descriptions. As an example of a current LCS we will concentrate on XCS [237] — not only because it is at the time of this writing the most used and best understood LCS, but also because it is in its structure similar to the LCS model that is developed in this book. Therefore, when discussing the theoretical understanding of LCS, special emphasis is put on XCS and its variants, in addition to describing general approaches that have been used to analyse LCS.

Even though the presented work borrows numerous concepts and methods from statistical machine learning, these methods and their background are not described in this chapter, as this would deviate too much from the main topic of interest. However, whenever using new concepts and applying new methods, a short discussion about their underlying ideas is given at adequate places throughout the text. A more thorough description of the methods used in this work can be found in a wide range of textbooks [17, 19, 102, 105, 164, 165], of which the ones by Bishop [19] and Bertsekas and Tsitsiklis [17] are particularly relevant to the content of this book.