Genetic algorithms for business optimization?
The best algorithm for optimizing behavior in a changing environment has existed for billions of years – The genes of living things! They have allowed living things to survive in ice, heat and all global disasters.
The gene sequences represent nothing more than a program of how a living being is to be built. Reproduction then provides everything needed for optimization and adaptation.
How can we use this algorithm for optimizations of IT systems?
Genetic algorithms – how can nature’s mechanisms be used for IT-supported optimization?
Nature uses the following procedure for optimization:
- Construction of a population with different genes
- Survival of the fittest – the fitter in the environment the more successful in reproduction
- Crossover: Mixing of parental genes by exchanging gene parts. This always happens when the egg and sperm cell are joined.
- Mutation: Random alteration of parts of a gene. In nature, this happens, for example, through radiation or also through incomplete repairs of the genetic material.
- Emergence of the new generation of children with the new, altered genes
Genetic algorithms mimic this approach.
Imagine there’s
- a program that can represent an enterprise environment or an enterprise goal
- and a program environment that can map the behavior in the corporate environment.
With this, we now represent a company optimization.
- Building a population: We let random programs emerge (hereafter called “individuals”), all of which represent a behavior in the corporate environment.
- Survival of the fittest: The programs are all valid and will better or worse achieve the goal to be optimized in the company. To measure this, the programs are executed and then tested in the corporate environment.
The more successful the program, the higher the “fitness value” of the program.
The higher the fitness value, the higher the probability that an individual will “reproduce”, i.e. that the program will be used for the child generation.
Example: Fitness value range is between 0 and 100. The probability that an individual with a fitness value of 50 will be used as a parent is then 50 times higher than an individual with a fitness value of only 1.
So 1000 parents are now being selected.
- Crossover: 2 parent individuals each are selected. A program section is now cut out of each and replaced by the other. This creates 2 new programs, the “children”.
- Mutation: With a certain probability, individual parts of the program are now randomly changed in the children.
- A new generation of children has emerged and the process begins again at step 2.
What is challenging about this approach is this:
- Find a good model for representing the corporate environment.
- Find the right fitness function and be able to define the goal correctly.
- Program individuals must be valid. A randomly built program must still work.
- Finding the right parameters: How strongly does an individual with high fitness prevail over an individual with low fitness? What is the mutation rate? What’s the population? After how many generations will the procedure end?
Despite the challenges, the benefits are immense. In the ever-changing corporate environment, Natural-AI finds a set of optimized solutions. Even those you would never have thought of, as diverse as nature.
TD Trusted Decisions remains at the forefront of this.
Interested in Natural AI optimization of your challenges? Feel free to contact us at office@trusteddecisions.com
