Pim van Meurs
Member
Member # 541
|
posted 31. December 2003 14:02
On ARN, Rock pointed out the following thesis by Mark Toussaint, also perhaps ironically, one of the authors of the No Free Lunch theorems
click 2003 The evolution of genetic representations and modular neural adaptation. link
quote:
For instance, we might have two organisms that look exactly the same (in section 1.2.4 we define precisely our use of the word phenotype) but if they are encoded by different genotypes, the probability distribution of their offspring may differ. How might evolution exploit this fact? Suppose evolution found an organism which is functionally quite well adapted but suffers from a lack of innovatability, i.e., all its children are no better than the parent. Now, evolution can change the genetic representation of this organism without changing its functional phenotype. This change of genetic representation, called neutral mutation, also changes the organism’s offspring neighborhood and in occasion will lead to more innovatability.
This innovatability or evolutionary plasticity or evolvability is a fascinating topic in evolutionary theory. Toussaint continues with the work by Schuster and Fontana which shows
quote:
The authors showed that the same phenotype (3D shape) can be realized by a large number of different genotypes. Depending on the genotype, the phenotypic neighborhoods change so severely that almost any phenotype becomes possible as an offspring when only a suitable genetic representation is given.
So in other words, although the same phenotype is realized through neutral mutations, selection may still play a role.
An excellent paper that ties together mathematically many of the aspects of evolutionary theory.
quote:
However, in natural evolution mutation operators are not designed by some intelligence. A central question arises: What does it mean to “learn” about the problem structure and exploit it? How in principle can evolution realize this? The answer we will give is that the implicit process of the evolution of genetic representations allows for the self-adaptation of the “search strategy” (i.e., the phenotypic variability induced by mutation and recombination). To some degree, this process has been overlooked in the context of evolutionary algorithms because complex, non-trivial (to be rigorously defined later) genetic representations (genotype-phenotype mappings) have been neglected by theoreticians. This chapter tries to fill this gap and propose a theoretical framework for evolution in the case of complex genotype-phenotype mappings focusing at the evolution of phenotypic variability. The next section lays the first cornerstone by clarifying what it means to learn about a problem structure.
[ 31. December 2003, 14:07: Message edited by: Pim van Meurs ]
IP: Logged
|
Printer-friendly view of this topic