Not Finding What You are Looking For?
   Tell Us!
(that's how we improve)

Unique Value

• Genetic Algorithms search large spaces of alternative solutions efficiently
• GAs are "robust", meaning they are suitable for a wide variety of problems because they make no assumptions about the nature of the fitness "landscape"
• GAs deliver good solutions quickly in large search spaces, but do not guarantee the optimal solution.  This makes them suitable for many applications where having a 99% solution now is better than a 100% perfect solution someday.

Description
Solutions are encoded as strings (chromosomes) of variables (genes) that have values (alleles).  For example, in the following graphic, the entire series is a chromosome, the positions are genes and the "1" or "0" values are alleles.  The entire sequence represents an encoded solution, or "geneotype".  When the genotype chromosome is decoded it becomes a "phenotype", a physical solution to a problem.  The performance of this solution is evaluated.  The performance measure is called "fitness".  The word "fitness" is based on how fit the solution is to solve the problem.

"Generation-based" Genetic Algorithms
In "Generation-based" Genetic Algorithms, a population of alternative chromosomes is created with random values.  These population members are decoded and their performance is evaluated.  Then using a performance-based selection scheme, the attributes of the better solutions are combined in a mating or "crossover" process.  Next, they are randomly "mutated" to introduce some "jitter" to explore the solution neighborhood.

The evaluation, selection/crossover, mutation repeats until some stopping criteria is met.  For example, one may stop after a number of evaluations, or a number of cycles, or when a certain performance has been found, or when a certain amount of time has been expended.

Continuous Genetic Algorithms
Continuous GAs do not have generations, but rather the population evolves on a continual basis as the algorithms are called upon to provide alternative solutions.

Like before, a random population is created, but when a solution is called for by the calling application, two parents are selected, mated and mutated and the child is returned.  That child is decoded and its fitness is evaluated.  If the child meets a performance criteria, it is allowed to return to the population.  If not, it is discarded.  Typically the child is better than the criteria and the poorest member of the population is dumped (to maintain population size) and the child is allowed to enter.  This is repeated while keeping the best solution found until some stopping criteria is met.

BioComp's GA-CPG algorithms are a member of the Continuous GA family.

The BioComp Advantage
There are a number of advantages in using GAs from BioComp:

• Nearly 20 years of research and development that has eliminated issues and tuned algorithms for fast convergence to a solution with few tuning parameters and insensitivity to settings.
• GA's that support constraints, limits and solution desirability
• Simplified GA technology with greater algorithm effectiveness and performance.  GA searches can be run with as little as four (4) API calls
• Scalable libraries are available that are being used in multi-threaded multiple simultaneous optimizations for large scale commercial applications
• Multiple-population optimization is supported enabling simultaneous search of significantly different encoding and convergence
• The technology is pre-embedded in useful applications for common applications
• The technology is commercially supported