What If Theory – AWIT

AWIT – Artificial Neural Networks What If Theory

Using an auto-encoder ANN  we will approximate the implicit function of any dataset during the learning phase and to assign a fuzzy output to any new input during the recall phase.

It is in this sense that we define AWIT (ANN What If Theory) as the interpretation of a dataset (traditionally referred to as the Testing Dataset) using the logic present in another dataset (the Training Dataset).

The algorithm to implement this new approach to data analysis follows:

  • Let DB1 and DB2 be two different datasets with the same types of variables but possessing different records;
  • The function f() is a non-linear function optimally interpolating DB1, by means of an auto-encoder ANN consisting of one hidden layer;
  • The dataset DB2 is rewritten using the ANN trained on the DB1dataset;
  • The output represents how each variable of the DB2 dataset is reformulated using the “logic” of the DB1 dataset.

References

[1] P.M.Buscema, W.J.Tastle
Artificial Neural Network. What-If Theory
International Journal of Information Systems and Social Change, 6(4), 52-81, October-December 2015. July 2015.

[2] P.M. Buscema, G. Maurelli, F.S. Mennini, L. Gitto, S. Russo, M. Ruggeri, S. Coretti, A. Cicchetti
Artificial neural networks and their potentialities in analyzing budget health data: an application for Italy of what-if theory
in Quality & Quantity, 10.1007/s11135-016-0329-y, Springer Science+Business Media Dordrecht 2016. 24 March 2016.