1
$\begingroup$

Why the circuit does not freeze/lock when the signals SUM ? But the electrical signals are SUM together in what sense?
This question is about neural network circuits because I don't understand what exactly at hardware level node is it
Is a line ?
I have a wrong image in my head about node concept because I translate like point:

if 10 machines meet at the same point or node occurs the incident, then how is adjusting the time within a neural network?

What happens inside a 'node' in artificial neural network ? Can you show me an example of hardware electronic circuit implementation about a neural network? a neural network?

Another suggest me this image to understand but I don't see circuit

a circuit with input terminals 10 to 10 independent signals.

But this is a line not a 'node'

I try to find a circuit for what you call 'node' about an artificial neural network. i don't understand how is implemented a 'node' with an electronic circuit

$\endgroup$
3
  • $\begingroup$ Voted to close because the question is unclear. It looks like a poor computer translation of some language into English. $\endgroup$
    – alephzero
    Feb 26 '17 at 21:12
  • $\begingroup$ I try to find a circuit for what you call 'node' about an artificial neural network. i don't understand how is implemented a 'node' with an electronic circuit $\endgroup$
    – lab teh
    Feb 26 '17 at 22:46
  • $\begingroup$ @labteh: digital, like in FPGA? Or analog? Analog node will be made with three op-amps, two to provide the gain with the tunable parameters, one to provide summation. But you won't find it in any practical use. $\endgroup$
    – SF.
    Mar 2 '17 at 17:26
2
$\begingroup$

First off, neural networks are only very rarely done in hardware, as actual summation of electric signal. Much more often they are just programs that run on standard computers, simulating a network with numerical inputs and outputs, and when you want superior speed and size, you implement it in hardware of FPGA, but still with numerical input and output. Analog neural networks, while possible and present in academic research, are not used in practice due to analog circuitry being considerably more expensive and bulkier than digital circuitry capable of simulating the same effects with adequate speed and precision.

And the actual summation takes form of $ax+by = z$ where $x, y$ are inputs, $z$ is output, and $a,b$ are parameters which are adjusted through the process of teaching the neural network, and since they can take negative values, $z$ may just as well come out as $x-y$ for $a=1; b=-1$.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.