What designs for genetic algorithms are there, if they are classified differently and/or have different names, that leverage models for epigenetics in evolution? What are the pros/cons of the designs? Are there vast insufficiencies or wide-open questions about their usefulness?

  • 2
    $\begingroup$ This seems a perfectly reasonable question to me - it's well defined and the answer within mainstream GA research is 'yes'. $\endgroup$ – NietzscheanAI Aug 4 '16 at 5:10
  • $\begingroup$ Not sure how the meta link or help recommendations apply; this is not a programming question but a request for high-level information about a specific area of artificial intelligence research. Edited it anyway. Hope this is a more useful question. $\endgroup$ – dynrepsys Aug 4 '16 at 14:53

Genetic algorithms are an analogy to biology, not a copy of it. The core piece of the analogy is that the "phenotype", or the observable portion of a solution, is constructed from the "genotype", or the internal portion of a solution.

For example, a number (the phenotype) can be stored as a binary series of "0"s and "1"s (the genotype), and by changing individual bits we make potentially dramatic changes in the resulting number, and by combining two genotypes we can get a broad range of 'related' numbers.

Epigenetics are wrinkles in the genotype -> phenotype mapping, making it a non-deterministic function, and so incorporating them would degrade the performance of a genetic algorithm by adding unnecessary noise.

| improve this answer | |
  • $\begingroup$ Ok, but if our genetic algorithms are far less effective than their analogs in biology, then isn't it reasonable for there to be efforts to improve the fidelity of the analogy? $\endgroup$ – dynrepsys Aug 2 '16 at 21:07
  • 1
    $\begingroup$ @dynrepsys: Are our genetic algorithms far less effective than their analogs? How would we know? $\endgroup$ – Matthew Graves Aug 2 '16 at 21:14
  • $\begingroup$ I guess the burden is on proving they are as effective, rather than proving the negative. Are there any genetic algorithms that are able to generate a surviving gene in an information space comparable to a small biological system over an equivalent number of generations with a reasonable cost function? I don't know of any but I'm open to the possibility that I'm just uninformed. $\endgroup$ – dynrepsys Aug 2 '16 at 21:33
  • $\begingroup$ "an information space comparable to a small biological system over an equivalent number of generations with a reasonable cost function?" - computationally that is HUGE. The biological space of a DNA genome is explored via physical function of proteins. Even calculating protein folding for a single gene is a supercomputer problem taking many hours of processing. And that says very little about its function in a chemical/physical environment, how it interacts with 1000s of other proteins in different concentrations. Software GAs are tiny search engines in comparison to biology of life. $\endgroup$ – Neil Slater Apr 12 '18 at 6:13

Over the last few years, evolutionary computation research has shown increasing interest in including some aspect of epigenetics. For example:

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.