Evolutionary Approach to the Implementation Problem
SEKANINA Lukáš. Evolutionary Approach to the Implementation Problem. Brno: Faculty of Information Technology BUT, 2006.
Evoluční přístup k problému implementace
Sekanina Lukáš, prof. Ing., Ph.D. (DCSY FIT BUT)
evolvable hardware, evolutionary design, implementation problem, Turing machine
Evolutionary circuit design and evolvable hardware traditionally belong to the area of electrical engineering. In this habilitation thesis, we interpret the evolutionary design of computational systems from the perspective of computer science. In particular, the evolution is able to produce a computing system satisfying the given specification; however, in general, we do not understand how and why the system performs a computation. It means that we cannot classify the evolved systems as computing mechanisms. On the basis of experimental results, we have shown the following: (1) There is no significant relation among the complexity of the evolved circuits, chromosome size and novelty obtained. Human-competitive results can be evolved at any level of interest if the problem is formulated in a clever way. (2) It is possible to speed up the digital circuit evolution using a common FPGA for a reasonable class of applications. (3) It is possible to evolve sequential circuits at reconfigurable platforms operating at sub-gate levels. In particular, the RS circuit was (re)discovered at the transistor level. (4) Evolution is able to form computational elements in extreme low temperatures. (5) The embodiment can be utilized to introduce additional functions to digital modules in a manageable way; in particular, polymorphic gate-level modules were discovered. All the problems investigated should contribute to the explanation what it means for a computational system to be designed evolutionarily and what we can expect from the evolutionary approach that works on the position of human designer.
Faculty of Information Technology BUT