2

I am working on series prediction by LSTM-RNN.

In the training stage, I use a random series (white noise ) as input to go through a system and get the output. LSTM is implemented to learn the relationsihp between this pair of input and output. I was wondering is this mapping can be learned by LSTM as it looks the input is random but the ouput is under some physical law.

And in the testing stage, I use another type of input, for example, sine wave. Can I expect a reasonable output via the LSTM-RNN model I built in the training stage?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Oscar
  • 21
  • 2
  • 3
    Possible duplicate of [How do I make my neural network better at predicting sine waves?](https://stats.stackexchange.com/questions/307210/how-do-i-make-my-neural-network-better-at-predicting-sine-waves) – Sycorax Aug 19 '18 at 15:31
  • 1
    LSTM, like other networks, can learn label distribution by ignoring input (white noise in your case) and utilising bias variables. However, such network would learn to predict most common label all the time which can be achieved by much simpler methods. – Tautvydas Sep 10 '18 at 12:30

0 Answers0