Genetic Algorithms in 3 Minutes

By ai-depot | June 30, 2002

Evolution & Applications

For Genetic Algorithms, the process used to simulate evolution has been simplified quite a bit:

Evaluation - For each solution, we try to solve the problem. It is usually then possible to judge how good it was: this is called the fitness. This is a crutial stage since this indirectly determines what solutions are good, and which are allowed to breed and which die out.
Selection - The selection phase picks two parents among the entire population, based on their fitness, the properties of the population and some random parameters.
Crossover - Given the two parents, their genotype is combined to form an offspring.
Mutation - The offspring’s genotype is then slightly modified, to provide more diversity.
Replacement - Based on the fitness of the offspring, we can decide which individual to replace in the population, if any.

Each of these stages has fairly significant impacts on the performance of the evolution as a whole.

Since Genetic Algorithms can evolve any genotype that can be evaluated, they are not limited to specific applications. So, among many others:

Constraint Satisfaction - Some computer science problems are based on variables with complex interdependencies, such as timetables. GA can be used to find options without clashes.
Numerical Optimisation - In engineering, some non-linear problems cannot - or take too long to be solved analytically. GA can speed up the process of finding solutions.
Robotics Controllers - Using simulators, it is often easier to evolve robotic controllers than to teach them by feedback in the real-world.
Artificial Life - More complex populations of individuals can be modelled and evolved with genetic algorithms. This falls into the Artificial Life category though.

Pages: 1 2 3

Tags: none
Category: tutorial |