Está en la página 1de 1

Special Forms of Evolution

Tarundeep Singh Dhot


Dept of ECE
Concordia University
Montreal, QC H3G 1M8

Coevolution and Interactive evolution are special forms of evolution as they


both work under external influence. In coevolution, this influence comes from another
population while in interactive evolution; it comes from a user (human) defining the
fitness values.

Coevolution is divided into cooperative and competitive coevolution. In


cooperative coevolution, increase in fitness of one individual increases the overall fitness
of the model (cooperation among individuals) while in competitive coevolution,
individuals compete against each other to gain fitness (fitness of one negatively affects
the others). Thus, cooperative coevolution can be seen as a mutual beneficial relation
(mutualism and symbiosis) while competitive coevolution is more of a non beneficial,
predatory or parasitic relation. Even though cooperative coevolution is dependent on the
user to provide a suitable partition to the problem, it permits effective function
decomposition as essentially each subpopulation solves a much smaller and amenable
problem. But the main issue here is deciding how to mutually pair solutions from the
subpopulations in order to improve overall fitness. Thus, the fitness landscape keeps
changing as different populations evolve according to the choice of pairing strategies. On
the other hand, in competitive coevolution, individuals compete against each other to gain
fitness at each other’s expense and are thus ranked according to the number of wins they
achieve (rank based selection). Eg: Iterative Prisoner’s Dilemma.

In comparison to coevolution, interactive evolution has a human as a part of the


system and evolutionary process. The influence of the human or user in the selection is
known as subjective or aesthetic selection. Thus, it suits situations having no clear fitness
function improving the search ability, exploration and diversity of evolution. But at the
same time it is slow, inconsistent having limited coverage. Interactive evolution can be
applied as an optimisation tool for problems having parameter based representations and
as a discovery engine identifying new designs as well as design principles analyzing
evolved designs for component based representations. A common example for Interactive
Evolution is the Mondriaan Evolver where a human is allowed to grade each image, thus,
being a part of the evolutionary design of the system.

Finally, we can see that in both Coevolution and Interactive Evolution, the fitness
that is awarded to a solution may vary as in Coevolution, the fitness is dependent on the
evolutionary state of the second population while in Interactive Evolution it is because of
user inconsistencies.

También podría gustarte