On the Question of Brain Activity as a Physics Problem?

[Delta1212]

What about remembering an image is more difficult to attribute to changing pathways? Image recognition is one of the first breakthrough areas that artificial neural networks were applied to in the modern era and achieved exceptionally strong results through the exact method that you're saying you don't think should work.

[Strange]

 

Because no one has ever proven that it's the wrong assumption to make

 

 

We can observe activity in multiple neurons at the same time.

 

 

it's impossible for two pulses start at exactly the same moment

 

That isn't what you said. That is a different (and as far as I can tell) utterly pointless claim.

 

As pulses do not take zero time, it doesn't matter if the start at "exactly the same moment" (which is meaningless) they will still be happening at the same time.

 

 

 

For example if a billion neurons fire in a second then how many fire in a billionth of a second. One.

 

That could (but not necessarily) be true IF a neuron could do anything in a billionth of a second. But of course, they take many milliseconds. So the answer to your question is quite possibly: all of them.

[Perfict_Lightning]

What about remembering an image is more difficult to attribute to changing pathways? Image recognition is one of the first breakthrough areas that artificial neural networks were applied to in the modern era and achieved exceptionally strong results through the exact method that you're saying you don't think should work.

 

Neural nets are great and have achieved a lot more then I thought they would. Here’s the thing, and please correct me if I’m wrong, but one of the best techniques to do the pathway reinforcement is backpropagation, and there is no physical basis for that in the organic system, unless you consider evolution.

In an organic system I can see that synapses that were particularly active would reproduce in number. The active neuron could trigger this, and that’s the whole mechanism behind reinforcing pathways in the organic system, I’m not sure if an inactive neuron looses buttons off it’s axon but I suppose it could to some degree, and worse.

With backpropagation in an organic system, it would be as if a neuron took an inventory of all the neurons it receives neurotransmitters from, and depending on what it wanted, selectively reduced buttons even if they were active, and create new buttons even with less active neurons. That kind of stuff only happens with evolution as the creature mutates. Especially, and here I’m not sure, with the degree of the reinforcement in neural nets. Like in an organic system one neuron may be receiving from 5000 buttons, but only 25 on a path to be reinforced, and they are active so are able to reproduce to 50 or 100, fine. In the neural net I’m thinking a rapid bump of 1000 in training wouldn’t be out of the question.

When a neural net is trained to recognize symbols, it’s like it evolved into the role, hard wired. Say it likes 7’s, it’s really good at finding them, can find them in the dark. But what about a neural net that can come in somewhat uneducated, see a few hundred symbols, and in each case learn enough during one exposure to be able to tell if ever sees that particular symbol ever agian. Do that without the benefit of backpropagation and use only changes in reinforcements of a similar level and timing we find in organics’. That would be interesting if it’s possible. I don’t know if it is but would look forward to what we could learn.

Check that result against how well an organic can maintain the new tissue over the life time of the organ. There is a lot of room but since there is a cost to doing the renovation, plus the cost of just firing and plus the incurred cost down the road, it just doesn’t strike me as though this is going on from a practical sense for the animal, or at least you’d see more of the control. If it was going to be expensive to remember photos then why am I so good at it all the time?

 

The reason remembering photos is more difficult to associate with reinforced pathways is because there is no readily apparent helpful pathway to reinforce yet we are so good at the task. However as you have pointed out the incoming signal does go through several hard wired stages so who knows. It could be there, but to be honest l still think neural nets simulate something closer to evolution, and in our brain today it’s the amount of new tissue it would require to be able to generate a large scale recollection that makes me question tissue as the source.

 

 

We can observe activity in multiple neurons at the same time.

 

 

That isn't what you said. That is a different (and as far as I can tell) utterly pointless claim.

 

As pulses do not take zero time, it doesn't matter if the start at "exactly the same moment" (which is meaningless) they will still be happening at the same time.

 

 

That could (but not necessarily) be true IF a neuron could do anything in a billionth of a second. But of course, they take many milliseconds. So the answer to your question is quite possibly: all of them.

 

 

Sorry guy, I think It’s a semantics thing, I’m just working on some robot software, the Silhouette Program, you might be interested in it,

Why define the moment the pulse starts as the moment a neuron fires?

Because you have to figure out which neuron is going to fire next. A neuron fires at one moment, fine, but at the same moment a parlif (as in the video above) is sent out from half the neurons. (There’s two sets of neurons, a right set and a left set and only one set sends out the parlifs). The parlifs grow and this moves them into the future, where they all look for active synapses. The parlif that is best able to find one, jumps the synaptic cleft, and the neuron that originally sent out that parlif fires. This connects the two moments in time, the moment when it fired and the moment when one of its’ buttons was active. The moments are connected while the parlif rides along length of the axon, storing the info it gathered about molecular locations along the axon. That means that at each moment in time a neuron fires and as well a cleft is jumped by a neuron.that had fired before.

For some of my inspiration, originally I wanted to use my eyes as a kind of biofeedback device where-by I could change which memories I was accessing by changing which direction I was looking, (an old NLP idea), and then I combined that with wanting to use information about molecular locations as memories. So in order to gather the information about where the molecules are at the moment, make a point, the origin, at the center of the nucleus of a neuron and draw a line from the origin in the direction you’re looking. Measure the locations of molecules in that direction. You can compare that with measurements made in the past. Now look in a different direction and you’ll get different molecules and different memories. So great you have some control over what you’re thinking. Now make a plane orthogonal to the line at that origin, and this plane doesn’t intersect with any molecules. Make a parallel plane right behind the first one with nothing in between. That is where we are going to store the information, between the planes, if we magnify it enough there is tons of room.

So all the other planes landing in the video to the left of the origin, are being stored between the two planes we made. There was more stuff going there then we thought and by the end the two planes aren’t so close together, they are about as far apart as the width of the synaptic cleft. At an active button, at one moment in time a synaptic vesicle breaks the surface of the presynaptic membrane, and at the same moment a molecule passes to the interior of the post synaptic membrane. Theses two events occur at the same moment and that is what enables the parlif to find the synapse and make the jump. The structures, because of their shapes and movements over time, are of great assistance to the parlif in what it’s doing, as are the rings of molecules that enter/leave the axon as the pulse moves along it.

The information concerning the locations of the molecules has been reduced to directions of planes as per the video. Imagine there is a sphere about the size of a small orange in the center of your brain. Represent all the consecutive moments in time much as you would a loose row of oranges. Now to map the information about the locations of molecules onto the pulse as it travels down/up the axon, associate with each moment between when the neuron fired and when it jumped the cleft, with an appropriate plane from the stack representing those locations. The first molecules hit are in erratic directions and so proly that end of the stack goes with the synapse end of the pulse. Now rotate each moment/orange to the appropriate direction as dictated by its’ associated plane. Those are like potential future moments and property of the firing neuron. Each neuron has its’ own such string of moments stretching into the future from when it fired.The actual present moment has perhaps the average tilt, the average of all the various representations ofthat moment held by the other neurons.

One thing, it could well be that finding out which neuron will fire next is as easy as throwing a dart at a dart board, hit any active synapse, follow the axon back in time and fire that neuron. But it is also possible there’s a 50/50 chance the active synapse was from the wrong set. One set of neurons sends out the parlifs to gather information, while the other set does hard storage back along the axon. They switch roles every time a neuron fires, so you see that synapse might not even be of the right set never mind the right neuron, and it all comes out in the wash, gets reorganized in a trip through the corpus callosum. Or maybe not and you can’t miss, I’m not sure, a work in progress.