Live data from Hacker News

Show HN: A Genetic Algorithm library written in JavaScript

github.com

1–10 of 42 posts

Re: Show HN: A Genetic Algorithm library written in JavaScript

#2
interesting.

I did proposed a GA framework in my final thesis, but I end up never actually sharing the code, because it the end you have to create most of the code anyways. How to create your population, how to evaluate each individual, how to cross them...

My final thesis is in Portuguese, otherwise I'll share a link

Re: Show HN: A Genetic Algorithm library written in JavaScript

#5
post #2

interesting. I did proposed a GA framework in my final thesis, but I end up never actually sharing the code, because it the end you have to create most of the code anyways. How to create your population, how to evaluate each individual, how to cross them... My final thesis is in Portuguese, otherwise I'll share a link

Do you have diagrammatic details to share ? That will be easy to understand irrespective of language.

Re: Show HN: A Genetic Algorithm library written in JavaScript

#6
post #4

Remember that great old flash thingy where you mutate a boxcar until it drives well, just found out there's a html version: https://rednuht.org/genetic_cars_2/

I made something similar for teaching shapes to walk [0]. It's actually a workshop to teach people at my job about how GAs work by having them implement the different parts themselves.

[0]: https://matsemann.github.io/walkingea/

Re: Show HN: A Genetic Algorithm library written in JavaScript

#8
Nueral network tech went through a weird period and then back to basics (back prop) just with massive computation (and simple activation functions) with fabulous results. I wonder if similar things could be done with other old school techniques like GA. Does anyone know if we have tried massive ones?

Re: Show HN: A Genetic Algorithm library written in JavaScript

#10

Nueral network tech went through a weird period and then back to basics (back prop) just with massive computation (and simple activation functions) with fabulous results. I wonder if similar things could be done with other old school techniques like GA. Does anyone know if we have tried massive ones?

There is plenty of work running evolutionary algorithms on GPUs, but there are a number of limiting factors that make it less than ideal. Evolutionary algorithms in general can be extremely effective, they just solve very different problems from "is this a picture of a cat?" and aren't as sexy as neural networks.

In fact, much of "old school" CS is still very much alive and applying modern techniques and hardware to their problems. It's not that neural networks are some field of CS that is much more advanced than the rest of CS but rather neural networks happen to be the current hot topic.

Post reply on HN