Notes on Evolutionary Algorithms

Posted on: Oct 8, 2023

 Initialization
       |
       |
       v
    Evaluate <-|
       |       |
       |       |
       v       |
   Terminate?  |
       |       |
       |       |
       v       |
   Selection   |
       |       |
       |       |
       v       |
   Variation --|

It is important to define a blueprint to describe our population

• Evaluation: the population is tested against a value and a fitness score is assigned Terminate: some criteria is defiend to decide if stop or not

• Selection: the population is selected to be used in the next iteration

• Variation: new population is generated e.g. by taking some strucutre to the selected population and mutating it (in order to not stay always in the same evolutionary line)

• Single Objective Problem: the population is tested against one objective. The solution is often trivial

• Multiple Objective Problem: the population is tested against multiple objectives. Can lead to a state where the solution is not obvious or trivial. An expert needs to visualize the results or an external criterium needs to be taken into account

Visualization via plot becomes hard when there are 4+ objectives to visualize. A solution for this is using “parallel coordinate plots”