Evolvability limits: a case study concerning the modular evolvable capacities (MECs) of a new neural net model for a second generation brain building machine BM2
Hugo de GARIS · 2003
This paper concerns a case study of the limits to which desirable properties can be evolved in a particular kind of neural network circuit (module). It has been the decade long dream of the author to evolve neural network modules in their 10,000s at electronic speeds in special evolvable hardware, and then to assemble them into a gigabyte of memory to build artificial brains. But such an dream is only realizable if the (modular) evolvable capacities (MECs) of such modules (i.e. a qualitative and quantitive measure of the quality of the evolution) are sufficiently high to make the effort worthwhile. This paper shows how the evolvable capacities of a module using a new neural network model (called DePo) was stretched to its limits. This paper makes the claim that evolutionary engineering is all about "pushing up MECs".