GENETIC PROGRAMMING Evolutionary Approaches to Multistrategy Learning

This chapter introduces the concept of Genetic Programming (GP) and dis­ cusses its relevance to Multistrategy Learning. Genetic Programming is concerned with applying Genetic Algorithms (GAs) to build/evolve complex systems. For the purposes of this chapter. complex systems are defined to be those that have struc­ tures and/or dynamics that are too complex to be predictable or even analyzable in practical tenns. With the coming of molecular scale technologies. it will be possi­ ble to build devices with a huge nwnber of components. The potential complexity of such systems will make traditional approaches to building and teaching them increasingly difficull An alternative approach is to imitate nature's method of cre­ ating (byper)complex systems. namely, a fonn of simulated or "applied evolution," which basically is what GP is about. This chapter shows how various learning strategies can be combined (e.g .• those from neural networlc.s and GAs or from cellular automata and GAs) to produce systems that are very complex in the above sense yet are capable of perfonning as desired By employing these multistrategy (in the above cases. dual-strategy) learning techniques. more complex and interest­ ing systems can be constructed than are possible by using the usual mono-strategy learning approaches. e.g.. neural network learning algorithms. such as "back­ propagation," or band coded reproduction rules for cellular automata