Having studied more or less in detail the mechanisms of working with the state-space that Mathematica provides us with, I asked myself: can these tools be applied to neural networks? This article helped formulate the problem:

On neural networks in identification and control of dynamic systems

There, in paragraph 3.2. on page 8, it shows what a neural network would look like in a state-space.

It is natural to assume that this will be a discrete state-space with the corresponding matrices A, B, C, D. But, how they should be formed, it is not clear ...

Thus, we need to connect all the elements into one system ("SystemModelMerge"?) maybe, get the state-space and convert from discrete to continuous form. I want to know what more experienced Mathematica users think and experiment with it.

Structure of neural network: enter image description here

  • $\begingroup$ Arn’t deep nn’s already a system with feedback? $\endgroup$ Apr 18, 2020 at 16:06
  • $\begingroup$ You mean, is this system not already represented in the state-space? If not, then clarify the question. $\endgroup$
    – dtn
    Apr 18, 2020 at 16:09
  • $\begingroup$ I’ll make a clarification, just in case. This is a feedback system. I am wondering how to represent it in the state space using Mathematics, i.e. get the matrices A, B, C and D. And convert to a continuous system. $\endgroup$
    – dtn
    Apr 18, 2020 at 16:13


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.