How to use Neural Network Error as a Feedback Input

Using neural network error as a feedback input helps reduce the overall network error and increase forecasting accuracy ( Wahheb et al. 2016).

How can I supply my Neural Network with its own error?

References:

Waheeb W, Ghazali R, Herawan T (2016) Ridge Polynomial Neural Network with Error Feedback for Time Series Forecasting. PLoS ONE 11(12): e0167248. https://doi.org/10.1371/journal.pone.0167248

Answer 1

[Waddah Waheeb]

The code to feed back network error as an input can be downloaded from the following link:

https://github.com/Waddah-Waheeb/RPNN-EF

Hope this helps!

[Greg Heath]

This is done automatically during training !!!

So...

WHY DO YOU THINK YOU NEED TO DO THIS ???

Greg

[Waddah Waheeb]

During training, errors are used to update the weights. But in the given code, the past error is used as an input too. Based on the literature in time series forecasting, this type of modelling is used to model nonlinear moving-average processes (e.g., unpredictable events or past shocks) more directly. Please have a look at this link.

[David Franco]

Thanks Waddah Waheeb! That's exactly what I needed.

Answer 2

[Greg Heath]

THAT IS WHAT HAPPENS AUTOMATICALLY WHEN YOU TRAIN THE NET ! SEE THE FIGURE

 net = train(net,x,t)
 figure

Hope this helps.

Thank you for formally accepting my answer

Greg