New evolution simulator video!

Author Topic: New evolution simulator video!  (Read 1257 times)

I started work on my evolution simulator for the first time in two months. Switching to neural networks I eventually get some interesting results:
https://youtu.be/faNIqivEufE


Very good to see some new progress :D

Very good to see some new progress :D
Thanks mate!

I'm thinking of adding a few different types of food (protein/carbohydrates) with different functions. Adding "radiation spawners" that let you change the mutation rate of certain areas, and adding back in loveual reproduction. Hopefully that will make things more interesting.

Very good to see some new progress :D
i guess you could say it evolved

What library did you use?

Wow, this is really interesting! Keep it up man, I really want to see what you can do, you seem very skilled.

So whats your end goal with this? Are you testing how best to create the ideal Übermensch?

What library did you use?
Sdl for events handeling, old opengl for graphics, liquidfun/box2d for physics.
Wow, this is really interesting! Keep it up man, I really want to see what you can do, you seem very skilled.
Thanks!
So whats your end goal with this? Are you testing how best to create the ideal Übermensch?
The new aryans will catch food pellets like no race before them.

Sdl for events handeling, old opengl for graphics, liquidfun/box2d for physics.
Huh, you wrote your own neural networks library? I thought you said that but I had to be sure.

Neat

Huh, you wrote your own neural networks library? I thought you said that but I had to be sure.

Neat
I mean it's not a terribly complex thing. You're just multiplying numbers and adding them together.

I mean it's not a terribly complex thing. You're just multiplying numbers and adding them together.
Cool can I see? Is it on Github?

I'm using neural networks right now with my Java project. Lots of fun but very stressful on the hardware and on my patience haha. Thank goodness for CUDA otherwise it would take me a month (per epoch) to train my neural network (seq2seq lstm with a very large bidi-map).

i'm glad you started working on it because it's amazing

Cool can I see? Is it on Github?
I haven't published source yet but I suppose I could.
I'm using neural networks right now with my Java project. Lots of fun but very stressful on the hardware and on my patience haha. Thank goodness for CUDA otherwise it would take me a month (per epoch) to train my neural network (seq2seq lstm with a very large bidi-map).
I always wanted to try using CUDA ever since I heard of it but I first tried it a few years ago when I couldn't even get the libraries/compiler to work. I am testing whether or not I got multithreading to work though.


i'm glad you started working on it because it's amazing
Thanks! It's always nice to know someone's actually interested!


Also, I tried to make a mini-map of the neural network. Except it's not that good because this is what *one* layer looks like:

That doesn't even take into account that the "genes" can also change the activation function of the neurons.