Neural networks, how to master them, where to start? - page 19

[Sceptic Philozoff]

It is interesting, for example, when we try to find s.c.a. returns on a large segment, considering it to be zero in the first approximation. The question is actually simple (let x be the desired series, N be the sample size):

True_co / Rough_estimate_co =

= MathSqrt( ( Sum(x^2) - Sum(x)^2 ) / Sum(x^2) ) =

= MathSqrt( 1 - N^2 * Average^2 / Sum(x^2) ) ~

~ 1 - N^2/2 * Average^2 / Sum(x^2) =

= 1 - N/2 * Average^2 / Average_square

Everything would be fine if it weren't for the multiplier N/2 before the second summand.

[Andrey]

It seems that in order for me to understand neural networks, I first have to learn the basics of some programming language

[Владимир]

Please help with the study of the Kohonen network. I don't understand how the number of dimensions of the hypersphere is found out. For example if inputs are 3 and neuron is 1, three dimensions? So 3 different inputs are coordinates of one point of one input vector? And these 3 inputs have to be linked by the same criterion? And if I need to add another criterion, create 3 more inputs based on it, and link them with the same neurons, would I get "unbalanced" measurements? The input vectors would be three dimensional, but the neurons would be six dimensional?

[[Deleted]]

danja >> :
Hi all! I am new here on a forum. Ran through a forum a lot of useful information:). I would like to ask the knowledgeable, I want to penetrate into the sense of neural networks, but do not know where to start while I have a general idea have read an article by Reshetov, now have a desire to develop in this direction. Popeopeye useful literature on this subject, maybe somewhere on a forum is, but I have not looked up? Thanks in advance:)

You have to start by understanding what optimisation methods are. And then it's downhill to the goals.