5 ms·

Hi, author here! I’m really excited by the conversation this has generated and I’m happy to answer any questions.

• vanderZwan 7y ago
  > When you're solving a problem, how do you know if the answer you've found is correct? In many domains, there is a single correct answer. A mathematical function may have a global maximum or other well-defined attributes. However, other problems, like how a cell behaves in a petri dish, do not have clear solutions. Enter evolution, which does not design towards a known solution but optimizes around constraints. What about situations when there are no real optima and minima, but only intransitive relations? So "rock/paper/scissor"-like scenarios[0]. Could you use genetic algorithms in any way in those circumstances? https://aeon.co/essays/attempts-to-choose-the-best-life-may-be-doomed-to-failure https://aeon.co/essays/attempts-to-choose-the-best-life-may-...
  • philipkiely 7y ago
    One issue with “rock paper scissors” is the fixed input domain; there is nothing to mutate if your three starting options are the only possible inputs. If you’re trying to do something like predict the next move based on historical patterns, I think a classifier would be better.
• ctas 7y ago
  Not sure if @vanderZwan meant the same, but what about scenarios in which your goal is to pick the best member of an already existing population where each member has a set of variables like agility, strength, speed. Would it be possible to find a pareto-optimal solution with genetic algorithms, maybe even with priority-parameters?
  • philipkiely 7y ago
    So one step of a genetic algorithm, selection, does what you describe, but the main value of the approach is the iterative generation of new populations.