What is a neural network, and can it be replaced with formulas?

Question

Machine learning

Do I understand correctly that a neural network is some kind of attempt to obtain some kind of non-linear dependence (but stable at the same time), based on a large sample (or small)? despite the fact that such a dependence, between the values ​​of the sample, is difficult to pick up manually. or am I wrong? then what are neural networks?

Answer 1

Well, unfortunately, for the sake of mass culture, a name was chosen that does not reflect in any way what it actually is or how it works. There are no neurons there and there is not even close, there is a logistic classifier that allegedly works like a "neuron" (which of course is not even close, it considers a simple formula, but the work of a neuron is a billion times more complicated), and a set of such logistic classifiers is called a neural network.
The work of such a network can not only be replaced by a formula, it is a formula. When learning, we only select the coefficients in this formula in such a way that it gives the most suitable results for us.

Answer 2

Well, yes, the neurons there are formulas like this:
neuron_output = [sigmoid from] (input_1 * weight_1) + (input_2 * weight_2) + ... + (input_N * weight_N) You
can make the outputs of other neurons as neuron inputs, which allows you to make the formula very large and multi-layered. Thanks to backpropagation algorithms, you can quickly pick up the weights without going through all the options.
I recommend that you first read about what an artificial neuron is, and then about neural networks
https://ru.wikipedia.org/wiki/%D0%98%D1%81%D0%BA%D...