Intrinsic emergence boosts adaptive capacity

Abstract

The essential part of research about evolutionary algorithms or any optimisation algorithm in general is the discovery of mechanisms to improve the search, when the problem is characterized by a huge search space. Many metaheuristics and hybridisations of those algorithms are invented and compared with their capacity to traverse this search space in the most effective way. One alternative approach when facing the problem of the space dimension is to discover some clever ways to reduce it. The notion of “intrinsic emergence” originally inspired by the developments of Crutchfield and Mitchell appears to be very helpful [2]. According to them and other authors, and as further discussed in [1], a macro property, labelled as “emergent”, should supply some mechanical and non-human observer with additional functional sense. This “functional device” replaces the common need of a human observation to characterize emergence. Indeed, as shown in [1], this concept offers an interesting way to encode macroscopically the genome of a multi-agent system, and by doing so, to reduce temporally the size of the search space. In [1], the different ways to observe the search space were tested randomly, while here, for greater coherence, the search in the “space of observables” extends the idea and uses an evolutionary mechanism. This combination of the two evolutionary searches is the core of the new algorithm presented in this poster. To test the effectiveness of this approach, the problem treated here is the evolution and the discovery of a cellular automata (CA) able to simulate a binary adder. This