How can neural networks learn functions with a variable number of inputs?

Question

A simple example: given the input sequence, I want the neural network to output the median of the sequence. The problem is that if a neural network has learned to calculate the median of n inputs, how can it calculate the median of even more inputs? I know that repeating neural networks can learn functions such as max and parity over a sequence, but only constant memory is needed to compute these functions. What if the memory requirement grows with the size of the input, for example, by calculating the median?

This is the next question about How are neural networks used when the number of inputs can be variable? .

Answer 1

If a neural network has learned to calculate the median of n inputs, how can it calculate the median of even more inputs?

First of all, you must understand the use of a neural network. We usually use a neural network in problems where a mathematical solution is impossible. In this problem, the use of NN is not significant / impractical.

There are other problems of this nature, such as forecasting, in which continuous data arrive over time.

One solution to this problem could be a hidden Markov model (HMM). But again, such models depend on the correlation between the input over a certain period of time. Thus, this model is ineffective for problems when the input is completely random.

So, if the input is completely random, and the need for memory grows

, .

: NN . , .

Answer 2

: . , , , . , , n , , 1/n. -, , , , .

Answer 3

: . N , n. 2 * N . N . n N, 0. N , . , 1 - , 0 - . Matlab: v - , 2 * N, v (1: n) . v (n + 1: N) . v (N + 1: N + n) , V (N + n + 1: 2 * N) . , . , , BCI.