The real life example 3
Identification of physical object code and dataWhen claiming new form of artificial intelligence, testing only data of dynamic systems would not be enough. This real-life example is taken from data published on OpenNN, more specifically from its GUI version Neural Designer.
The both OpenNN and Neural Designer applications are accompanied by multiple experimental data for training end users. One of them is Airfoil Self Noise. It is formatted as symbol separated rows, first five are inputs and sixth is a target. This is how data look like:
800;0;0.3048;71.3;0.00266337;126.201 1000;0;0.3048;71.3;0.00266337;125.201 1250;0;0.3048;71.3;0.00266337;125.951 .....................................According to training and testing procedure provided in Neural Designer help the entire data set of 1503 records is divided into training part (60%), selection part (20%) and testing part (20%). The training part is used for building multiple neural networks. These obtained models are tested on unseen selection data. The best for selection model is used for testing on testing sample. The target computed by model is compared to actual target and Pearson correlation coefficient is computed. Neural Designer help reports 95.2%, in our experiments we obtained same 95.*% as the best result for 10 models using Kolmogorov-Arnold representation.
