Neural Nets simulation

[ACULTURE0]

Efficiency... do you see?

[7Stark]

would look good as a screensaver

[Longlivethe4th]

the layout of the network kind of reminds me of the galaxies in the first Elite game...

[rynjahninjah]

type won before youtube!!! its pretty cool

[BlazeCyndaquil]

Quite interesting. Did you make this program. If not, do you know where I could obtain it? I am assuming you did program it though. If that is the case, would you mind posting a link for it or something?

[reexre]

Please could you tell me if this is correct?

watch?v=sPHwPbGP5Pk

somthing to improve?

[mission2ridews]

Did you just hard code the graph or did you happen to make an interactive network builder?

[mrverde8080]

fake

[Houshalter]

Whats the point of a neural network. Is it faster or something. I don't see what the advantage is to use one in any situation.

[badmephisto]

neural networks have a lot of nice properties. They are general function approximators and yes they are pretty efficient most of the time. Most recently there has been some impressive progress in deep belief nets and convolutional RBM's that looks nice

[Houshalter]

@badmephisto, ya, but how do they do boolean logic functions, and even if they can, they have no memory.

[turoniable]

@badmephisto If I remember correctly the study of neural networks had the output of the oop programming.

[Bluebull114]

Text recognition, face recognition, giving computer intelligence, processing any visual data, that cannot be processed "the old way".

Search for openCV.

[Zoiros85]

@Houshalter Thats like asking whether brains have any advantage. What can you do better than a normal computer system?

[Houshalter]

@Zoiros85, alot of things, but if my brain was simulated on a computer it would defeat the purpose.

[Zoiros85]

@Houshalter Why would it defeat the purpose? Then you could have several of you solving problems. Anyways, in short: Neural nets can approximate complex functions very well and do it without a thorough description of the problem. This is in contrast to many other methods like decision trees which are generally faster learners.

[Houshalter]

@Zoiros85, ya, but the only advantage a neural network has is that its massively parrallel consisting of possibly billions of different simple componets all interlinked. If you run it on a serial machine, it loses that advantage and bogs the computer down with countless many small calculations, many of which are useless or unecessary.

[Aoitetsugakusha]

@Houshalter That's true with respect to speed of computation, but not the only advantage. Even if the serial computer is only 'simulating' the network, it's still performing computations the same way, mainly slower. Neural networks are also robust to problems faced by other controllers, can be scaled up in power and size quite easily if implemented right, and in general are good scientific tools for studying cognitive systems (that last one I'm talking from experience).

[hyperthreaded]

So, what is it thinking?

[sea4ever7241]

What makes them fire? Is it based on thresholds? I've only read about feedforward networks, or is it a spiking network as mentioned in a comment above?

[greenheadjoe]

awesome

[scorpius420]

also what programming language? Many thanks for any help! :)

[badmephisto]

Its made in Python.

It is NOT good, and was made looong time ago. I have since developed a MUCH more biologically correct model. Maybe I will post it sometime soon... the new one Does have Hebbian Learning, and also stuff like Inhibition, Refractory periods, and it Can compute output for set of inputs and do recognitions.

[scorpius420]

got it...sent you a message. Great work dude.

[kscrosson]

Please post!

[laputahayom]

where can i get it?

[Vigrante]

@badmephisto

That sounds really awesome. You need to post it haha. I would also be interested in the source code.

[TheReasonWhyGuy]

Did it died?

[RLore18]

That's not a simple one-time feedforward network. That's a "massively parallel" network, where the neurons fire at different times dependent upon input and/or time intervals.

[badmephisto]

Well I never heard it in those terms. If you want to get technical it is simply a recurrent spiking neural network...

[scorpius420]

does it just use thresholds and weights or does it have any short term synaptic plasticity or facilitation or other such non simple summative+threshold behaviours?

Does it do a task? did you try training it?

[c0d37rulez]

this video show how virus infect network xD

[wertstahl]

This is actually giving me a very pleasant feeling during watching... It was sort of the kind of movement and spread of movement i like to see. Funny..

[plmqas]

It's nice video Badmephisto, I'm very interesting which program or computer language did you use to write this ?

[juancho179]

Great thing... Just such as biological neurons.

[artilekt]

very interesting. What is model of one neuron? It looks like quite different then clasical neural network, because you used time.

In human body there is time too.

[badmephisto]

yes these are called biological neural networks

[superscatboy]

That looks less like a traditional computer neural network and more like an attempt at simulating a biological one, but nice work all the same.

[badmephisto]

yes... that was exactly the point... in the description I also say "this is what i imagine our brain cells work like".

[superscatboy]

That's cool. Have you trained the net for any particular purpose yet?

[badmephisto]

no... i haven't figured out 1. how to train that thing and 2. what to train it on. 1. in particular is casing me headaches :) Perhaps a genetic algorithm? but genetic algorithms are not too useful in chaotic systems, and i would be inclined to think that this one is. i dont know

[superscatboy]

I would say try a GA, but I would say that - I love 'em! ;) I recently discovered Pyevolve, a GA library for python that really does all the hard work for you - all you really need to do is write a decent fitness test. There may be a similar library out there for whatever language you're using.

[badmephisto]

I looked into pyevolve, it looks awesome! Will definitely check it out in more detail soon, thanks.

[TheBestNameEverMade]

I imagine its like a neural network. Overtime if connections produce bad results they send a smaller signal until its pretty much zero.

[masterman321]

FIREWORKS...yay

[kelt3we]

What kind of net is this?

[SonnyKraack]

its a model of nervecells.

[rbolo29]

Anyone know if the keys to Artificial Intelligence may lie within "FRACTALS?"

[davidwizard2006]

if that staff has assymetric property, ai is doomed to fail

[IrreducibleParadox]

If all of intelligence was asymmetric it could still potentially be digitized but it would be a lot more like capturing a soul. :-p

Beyond methodology.

[blytqb]

very good! i'm tinkering with NN.. was writing a framework in Java, but now i'm learning Python as fast as i can to write a more general framework, which is like a multi event architectur: you can build i/o blocks with a specific event signature and link them.

..Anyways I'm impressed with visualization.. any info about it???

[eelkonio]

Great visualization. Reminds me of my university days, when I complained to the professor that the NN models he taught didn't take into account the time a signal needs to reach another neuron. The true 3D location of a neuron needed to be taken into perspective, I urged.

But no we shouldn't look at that, he said. It was of minor importance, he gathered. Wrongly, I still think...

[badmephisto]

I absolutely agree with you, and I think that is one of the biggest downfalls of current NNs maybe. The brain uses a substance called white matter to coat the connections between neurons, and in fact the the thickness of that film regulates the speed of impulse that travels down that connection. So the brain has a way to regulate the speed, and it is most definitely a very important factor. Neurons have also been known to participate in well timed 'orchestras' of activity too.

[eelkonio]

I'd go even further with this. Mobile (!) glia-cells are highly involved in the brain.

I've derived that glia-cells -of which there r 10 times more cells than there r neurons- r very much involved in communication networks inside of the brain.

It turns out they can communicate between themselves and even order their neighbouring neurons to fire! Glias also determine which neurons should live or die. They work local, but can do communication to other groups of glia-cells through neurons.

[badmephisto]

I always thought of the glia cells to be just matter that holds neurons in place, insulates them, and provides nutrition. I'll have to look more into those I suppose :)

[eelkonio]

Oh, definately! They don't talk in zeros and ones like neurons - they excrete chemicals in certain amounts which produce 'waves' of communication. When it goes awry, you've got a case of epilepsy on your hands: calcium waves excreted by glia-cells.

Glia cells are by far the most intriguing and most ignored part of the brain.

A study in december last year found that glia-cells control the sexual preference of fruit-flies. There's something to think about. (See Nature Neuro magazine)

[johnnyk427]

That explains why I'm sexually attracted to fruit flies: too many glia-cells.

[NPderive]

It's not a matter of how many glial cells (similar in each individual), but rather the size of the glial cells, as they envelop the neurons.

[plmqas]

eelkonio it's very good that you wrote about this, because, the role of glia cells was highly underestimated and that discoveries are quite new and aren't wide-spread, discovery of their rules, should change the way of building optimalization algorithms which optimize artificial neural networks with undetermined, a priori, architecture/topology

[agenttachyon]

Like a pulsing neural net?

[virgildisgrace]

What is the visualization programmed in? Are you actually using weighted sums?

(also: death? uh.. what resemblance does this bear to "death"?)

[badmephisto]

its in python. and yes i am using weighted sums

[stompy243243]

Cool beans!

[infinitee44]

reminds me of death.

thanks!

[inbox101]

.... its actually a theoretical model of your brain...

[infinitee44]

I understand, but is that a valid reason Not to associate this video with the concept of death?

The fractal pattern seems to be everywhere you look in the world. Microcosm, macrocosm - an archetypal pattern for living processes, perhaps?

[reiner7674]

wonderful thanks for uploading