In the direct nervous network, can I turn on the “don't care” input function?

Question

I created a direct neural network using DL4J in Java.

Hypothetically and so that everything is simple, suppose that this neural network is a binary classifier of squares and circles.

The input, the vector of the function, will consist, for example, of 5 different variables:

[number_of_corners,
number_of_edges,
area,
height,
width]

So far, my binary classifier can distinguish two figures from each other, since I give it a full vector of functions.

My question is: is it possible to introduce only 2 or 3 of these functions? Or even 1? I understand that the results will be less accurate in this case, I just need to do this.

If possible, how?

How can I do this for a neural network with 213 different functions in the input vector?

Answer 1

, , area, height width ( number_of_corners number_of_edges).

, , , 10 10 , 10 , , number_of_corners number_of_edges. 10 ( ).

, area, , area, - . (I.e. area , .)

"", , () .

, . , , Deep Belief Network , , . , DBN "" ( , ); .