1
$\begingroup$

A am interested in physiologic neural network. Altough there are some opposite views, most probably there seems to be no plausible way to explain a physiologic backpropagation in the brain.

So I am trying to code a neural network without backpropagation yet my mathematical understanding is inadaquate, so I want to ask folowing simple question:

“If we do have only one node at the right side of the network, accepting that the inputs are on the left, can we train the network without backpropoagation and using the mean of weights instead? As we would know the y for all x it should be possible to calculate the mean w?”

The idea is that the system should work continiously, and train continiously for one class, until the trained network decides that the input is different from the known previously trained classes. And if it is different, it should create and train for that new class of inputs. I think that should be the working system of the brain, and as the cortex has similar cells, mathematically it also must be that easy?

But where is my flaw (with simplified math please :))

$\endgroup$
13
  • $\begingroup$ If there is only one class, why do you need a neural network at all? Just do function getClass(input) {return "class 0";} $\endgroup$
    – user253751
    Mar 23, 2022 at 11:30
  • $\begingroup$ Thanks for the answer. $\endgroup$
    – user53603
    Mar 23, 2022 at 12:12
  • $\begingroup$ Thanks for the answer. True. But would a training result in a mean of weights? If yes, then it would be able to differentiate that class 0 from others, that are significantly different from “other” classes.Than it would begin to train for that new “ class”…So we would reach a new class producing system or general intelligence without backpropagation? $\endgroup$
    – user53603
    Mar 23, 2022 at 12:19
  • $\begingroup$ If there's only one class there is nothing to differentiate. If you have "class" vs "not class" that's 2 classes $\endgroup$
    – user253751
    Mar 23, 2022 at 12:47
  • $\begingroup$ I am really sorry for my childish question, but I do want to understand: If we know the range of x that would be a cluster of information, and classify that information as class 0, wouldnt that be still one class to train? $\endgroup$
    – user53603
    Mar 23, 2022 at 13:38

0

You must log in to answer this question.