Genetic algorithms, what's the point?
thoughtless shitposting, what's the point?
life, what's the point?
replying, what's the point?
memes, what's the point?
Proprietary software, what's the GNU/point?
Maki, what's the point?
>was going to post about genetic algorithms
>then I saw it was a meme thread
Fuck all of you, I got excited for this.
>>53584228
>>53584270
these are the only two that do not have real purpose.
>>53584228
>Genetic algorithms, what's the point?
a cure to africans
>>53584228
>Genetic algorithms, what's the point?
They work.
>>53584228
How does the wizard install windows software from a macbook?
>>53584440
he uses teamviewer to install it on your computer. what would be the point to install it on his computer?
>>53584328
That's a shop my friend
>>53584440
he's a fucking wizard anon, thats his thing
I was hoping for a cool thread on GAs in here... oh well
>>53584228
They work pretty badly if the search space is smooth, you should only use them if you have many steep local minima
>>53585450
No that is a car.
This is a shop
>>53584228
They are a global optimization method. Given enough time GA will find the genetic sequence that globally maximizes the fitness function.
>>53584340
Don't go yet, anon. I may talk a bit about my GA projects.
>>53587119
nah. depending on how you do things, your population or populations can well converge to one or multiple local optima and stay there. as always with global optimization, you can't guarantee shit.
>>53584313
saged
>>53587130
This senpai, I recently started working with the Neuroevolution of Augmented Topologies and it is really interesting :)
>>53587702
Oh hi, another NEAT fan. Interesting stuff indeed, a shame it doesn't appear to be useful performance wise. I tried implementing it for massive parallelism on GPU once, but unfortunately failed miserably.
>>53588084
Oh wow that actually sounds really cool, care to share the experience?
What puzzles me about genetic algorithms is that the entire basis (evolution) is just a good theory at best. Really seems like a waste of time to replicate something so shaky.
>>53589250
>theory
>a guess
/sci/ pls go
https://youtu.be/YZUNRmwoijw
>>53584228
Genetic algorithms is when RNG writes better code than you.
>>53588977
Not much for a story I'm afraid. I was taking a parallel algorithms class based on OpenCL. Grades were based on a two-week team project, and together with two friends I decided to try this. We came up with an architecture that implemented mutation, crossover, and fitness evaluation as several OpenCL kernels (genetic operations parallelized at the level of individuals, network evaluation prallelized at the level of neurons). We started hacking things together and, two weeks later, had something that at least seemed to do "something". We didn't have the first idea on how to verify that the genetic operations were running correctly, but, given that we had no success in making the thing learn something (basic benchmarks like xor), I very much doubt it. To make it worse, our parallel network evaluation was slower than what I had coded ad-hoc in Java for comparison. And that was the end of parallel NEAT. I still like the idea and suspect that, given enough time and competence in parallel programming, something might come from it, but unfortunately I have neither one.
>>53589631
Ah. Well thanks for the story mate. I say if you ever get the change you should keep on going because it is certainly an interesting project and definitely worth while for investigating. Maybe you could even write a paper/publish it :^)
>>53589250
8/10 original context
>>53589661
Yeah, back when we did this neural networks were still written off as a nice but ineffective idea from the 60s. Given the current hype, especially when you combine the terms "neural net" and "GPU", it suddenly appears in a much more attractive light.