TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Ask HN: Can genetic algorithm get “sick”?

5 pointsby tuxlinuxienabout 9 years ago
ok, my question might be stupid but I was wondering if my genetic algorithm get "sick", how can I find a cure for it?

3 comments

dwils098about 9 years ago
By sick, you mean can it become bias to a subspace in the solution space, if so you would "cure" it by re-defining your fitness function to not get "stuck", or fooled by a local maxima.
crazypyroabout 9 years ago
I don&#x27;t understand what you mean by &quot;sick&quot;. There are a few different ways a genetic algorithm can &quot;fail to converge&quot; though (or converges on a solution that is not a global maxima).<p>This could be caused by having too high of a diversity, causing the algorithm to give up on promising avenues of evolution too early.<p>Another risk is having too high of selective pressure, causing the algorithm to get stuck on a &quot;good&quot; solution, but not a global maxima (best solution). A high selective pressure will reduce the total breadth of the search and create a deeper, more focused search on a smaller area (in general).<p>There are other factors to consider, such as bloating (esp. in genetic programming, where syntax trees can become unmanageably large) and the need for an accurate &quot;fitness&quot; function, less you select for something other than the actual solution.
pizzaabout 9 years ago
You may want to look into poisoning attacks or the no-free-lunch theorems for interesting technical aspects of reinforcement learners