[danlovy]

Greetings,

I've just completed phase one of a project to create neural simulators using genetic algorithms.  Video is here:


Source and discussion is here (still filling this in, but source code is available):
danlovy.com/critter

Would love feedback etc.  Trying now to figure out where to take this.
I have a number of ideas but wanted to get input from the larger AI community.

Best,
Dan

[keghn]

 Very interesting. I could take this to human level. But that is a lot of work and may years. I in for the ride.

[danlovy]

The real challenge is bringing industrial scale.  That's about process.  Is it possible to create a platform that can leverage multiple contributors effectively?  There are plenty of examples of genetic algorithms being used to simulate neural behaviors.  The problem is that they all stop at the application.

I would propose a general solution that builds off previous generations (what I was demonstrating) and allows organic growth through group contributions.

Sound right?
Dan

[keghn]

 I have given thought to similar project of mine.
 In the neural network world they dived artificial neural networks into to main type the
detector, like a detector of food, taste or pixel detector.
 The other part is the generator Neural network that generate something from the activation of a detector. Such
as motor movement to the food or away if the light is to bright. What is interesting to me if this develop in real organic system or
if it can be done in organic systems?
 I sure in early life the two would be closely entwined?

[danlovy]

I agree.  What I want to create is a general platform that can take contributions from others.  I am trying to figure out how to organize this.

Dan

[keghn]

 I like if there was a web page of example that copy right off the page. Then pasted into a text file then compiled them.
 Like here for example:   
http://www.seriss.com/people/erco/fltk/

 A two directional approach could be used?
 First is to state from scratch. The second would  be a complex model with soft switches
already set. So that when it is ran person would just need to make minor changes to get what the want.

 For example. I wanted to do the dynamics of a air foil in computer fluid dynamic simulator. But i have to build up from
scratch to get something working at sell level. Many and many soft switch have to pushed to get it work at sea level. A
permutation wall to stop new people right in there tracks.
 I would like just have a complex model already running at sea level, at a set speed, and set air foil, with easy to use dial on the
side to make minor adjustments to speed and air pressure.

[ranch vermin]

I think if you did it properly, you could reconstruct that little critter,  its a conspiracy that it hasnt been done before.   But im not into bio-tech, I keep everything mathematical and away from nature.   Because in my mind its morally wrong to do,  other scientists that come from a different viewpoint would be different.

[danlovy]

Ranch Vermin - what are your moral objections?  Given those objections, what is your interest in AI?  Seems like a contradictions

[ranch vermin]

Theres a clean way to do ai,  and its the opposite of being like that Jeff Hawkins guy.  im not strapping monkeys to chairs to do something which i think is simple enough to not bother with that kind of thing ever.

Fast random access networks and quantum mathematics generally have very little to do with biology unless u want to draw a course similarity.


I know that a fly can piss you off enough to make you think of reverse engineering his brain, but i could never follow through with it.

other scientists,  go through a set of pork chops and wonder why i worry so much.      but im not like that, and i never will be.  but i guess in the carnivorous environment we are in the whole things a contradiction.

just imagine being the animal your doing it to, as you do it.

[keghn]

  In my neuron model i use neurons in a daisy chains. A daisy chain of 7 neurons in a loop would be like a delay line.
 In my neuron model uses spiced pulse move through neurons at a rather slow speed. Artificial Neurons are near the speed of
light and would need a million of artificial neurons in a loop to hold one pulse. Bust spiked neuron being slow many pules
con be placed within a loop of neurons. To work as a memory cell:

https://en.wikipedia.org/wiki/Delay_line_memory

https://www.youtube.com/watch?time_continue=1&v=N9cUbYII5RY 

 A binary counter can be be made with the fist having a loop of two neurons and the the next loop would be made of four neurons and
the next would be eight neurons, and so.....

 So with daisy chain of neurons could play video. A pulse travel forward through a daisy chain neuron or binary counter. And a second
line of comes off the side of each neuron and goes into Deep generative that play video on the output:

http://blog.otoro.net/2016/04/01/generating-large-images-from-latent-vectors/ 

 So My nero brain model is a detector. Then a daisy chain of nerves to guide timing steps.Then a generator NN to apple the right action.

 The basic design of my neuro brain is to have a detector NN. When it detect food or a letter it output to a location on a very long brain
 daisy chain. That caused a spiked pules to come into existence. The pules coming off the daisy chain goes into Generator NN to
lay video sequence or muscle sequence open a clam shell. 

[danlovy]

Well, Episode 2 is released.  Now the creatures can move so the animation is much more interesting.

[keghn]

 
  What is next 3 d FPV space?

[Art]

Hello Danlovy,

Welcome!
I didn't actually see that you had joined our Forum and certainly did not want to ignore your presence.

Hopefully, everyone will benefit from the informational exchange, no matter great or small. Please enjoy your time here!

- Art -

[Zero]

I agree.  What I want to create is a general platform that can take contributions from others.  I am trying to figure out how to organize this.

Dan

Welcome danlovy! You could make it run in browsers web workers. This way, contributors would just have to visit your website to contribute.

I remember a nice language called Push3, intended primarily for use in evolutionary computation systems.

[danlovy]

I agree.  What I want to create is a general platform that can take contributions from others.  I am trying to figure out how to organize this.

Dan

Welcome danlovy! You could make it run in browsers web workers. This way, contributors would just have to visit your website to contribute.

I remember a nice language called Push3, intended primarily for use in evolutionary computation systems.

Thanks.  I chose C++ because I'm old.  I haven't done any coding in many many years.    It's the language I'm the most familiar with.  I was torn between that and JavaScript.  I could see the advantage of simply running in a browser but I wasn't quite ready to commit to the learning curve.  This is an experiment and the next version will have a language decision process.  If anyone would like to take on a JavaScript refactoring...