Background process

• Population size = 1000 agents
  

  training set = 25 agents:

  {

  15 fittest agents from foreground 10 fittest agents from background

  
  
• Iteration:
  1. All 1000 play against each 25 in training set
  2. Fitness evaluated
  3. Best half mate and generate new bottom half
  4. Best 10 (with fitness sharing) go to training set
• Every foreground iteration:
  1. Population is reset
  2. New 15 best are received