Is a 2D brain possible?

[fredreload]

Is it possible to create a brain in 2D based on circles and lines like a decision tree?

[Strange]

I doubt it because you wouldn't be able to get the level of interconnectivity needed. We can only make very simple electronic circuits in 2D.

[fredreload]

I doubt it because you wouldn't be able to get the level of interconnectivity needed. We can only make very simple electronic circuits in 2D.

Imagine flattening your brain, you still have all the connections, they just overlap each other

[Strange]

Imagine flattening your brain, you still have all the connections, they just overlap each other

 

OK. If you don't really mean 2D but just very thin (like an integrated circuit) then yes. But in 2D the connections don't overlap they short-circuit.

[fredreload]

 

OK. If you don't really mean 2D but just very thin (like an integrated circuit) then yes. But in 2D the connections don't overlap they short-circuit.

How about spread it out so they do not overlap like a puzzle, as long as the color moves on top it resembles the flowing current

 

P.S. Hmm, that brings possibility to a 1D brain

Edited July 29, 2016 by fredreload

[Strange]

How about spread it out so they do not overlap like a puzzle, as long as the color moves on top it resembles the flowing current

 

There is a limit to how many connections you can have without crossing.

[fredreload]

It doesn't matter, when an electrical pulse goes out, you just manually create a current at beginning of another line's starting point. --------* jump -> *----------

Although I'm not sure if you can create an actual computer with it because you'll have to swap bits

 

P.S. But still imagine connecting the entire nervous system to a single line and just loop the current at various points of a line

Edited July 29, 2016 by fredreload

[Strange]

A jump implies a third dimension.

[fredreload]

A jump implies a third dimension.

That's tricky =/, does that apply for the color moving as well?

Alright then Strange, they are now circles that make current loop, many many circles, no jump, beat that

Edited July 29, 2016 by fredreload

[Strange]

So you just have a number of independent current loops. What use is that?

[fredreload]

So you just have a number of independent current loops. What use is that?

How about a cog type of design? You swap current(wood) to the other cog, no jumps. Right you still need to swap it. Alright I guess you need to exert some type of control and you're saying this control is 3rd dimension. How about having it diverge? One loops back the other goes forward, no control there

Edited July 29, 2016 by fredreload

[Strange]

How about a cog type of design? You swap current(wood) to the other cog, no jumps

 

An interesting idea. If you can design a Turing (compatible) computing machine based on that then, yes you could build a brain!

 

You could use some of the circles for storage and some for communicating data from one place to another. You would have to work out a way of making basic logic gates or other decision functions. But it might be possible.

[fredreload]

 

An interesting idea. If you can design a Turing (compatible) computing machine based on that then, yes you could build a brain!

 

You could use some of the circles for storage and some for communicating data from one place to another. You would have to work out a way of making basic logic gates or other decision functions. But it might be possible.

Well a computer requires inputs and like you said, inputs are 3rd dimension

Well I've argued in another forum that you can't simulate a computer faster than the existing computer's hardware build so

[ydoaPs]

Sure you can. A simple Artificial Neural Network (even Deep Neural Networks or anything that isn't convolutional) is 2d. You have a layer co-ordinate and a neuron co-ordinate.

 

And if you don't want it to be software, you could train the software neural net to your liking (make sure the neuron outputs in the input layer and hidden layers are binary [so a step function instead of a sigmoid or tanh], though). Then create a conversion between weights and resistance and between input sum and capacitors. Each capacitor (and a resistor to standardize the outputs)acts as a neuron, and resistors of various resistance act as the weighted connections.

 

I'm not sure why you'd want to do that, but you could.

[fredreload]

Actually you are creating multiple timers with trade off on collision detection, but then you need to take into the account of parallel computing , it might work

P.S. Actually, you can have infinite timers, but you still need collision checks in the subsequent circuit design, which would take a good few if statements, depends on how many of that you can run in parallel

Edited July 29, 2016 by fredreload

[ydoaPs]

Actually you are creating multiple timers with trade off on collision detection, but then you need to take into the account of parallel computing , it might work

 

P.S. Actually, you can have infinite timers, but you still need collision checks in the subsequent circuit design, which would take a good few if statements, depends on how many of that you can run in parallel

Yeah, I didn't think about them just pulling charge until they go. We'd need a way to discharge them outside of the network upon a firing of a neuron in that layer. Maybe each neuron gets a set amount of time (like 1/4 second or less) to fire, otherwise it discharges to someplace outside the network?

[Strange]

Well a computer requires inputs and like you said, inputs are 3rd dimension

 

You would keep the inputs and outputs around the edge so it remains 2D.

 

 

Well I've argued in another forum that you can't simulate a computer faster than the existing computer's hardware build so

 

Indeed. Depending on the level at which you simulate, it may be hundreds to millions of times slower. This is quite a challenge in microprocessor design. We sometimes use large numbers of computers networked together to very slowly simulate a few milliseconds of execution.

[Delta1212]

Sure you can. A simple Artificial Neural Network (even Deep Neural Networks or anything that isn't convolutional) is 2d. You have a layer co-ordinate and a neuron co-ordinate.

 

And if you don't want it to be software, you could train the software neural net to your liking (make sure the neuron outputs in the input layer and hidden layers are binary [so a step function instead of a sigmoid or tanh], though). Then create a conversion between weights and resistance and between input sum and capacitors. Each capacitor (and a resistor to standardize the outputs)acts as a neuron, and resistors of various resistance act as the weighted connections.

 

I'm not sure why you'd want to do that, but you could.

 

You can't physically build a neural network in two dimensions, though. The connections would intersect.