Employing nonlinear transformation of datasets to train neural networks

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F47718684%3A_____%2F24%3A10002350" target="_blank" >RIV/47718684:_____/24:10002350 - isvavai.cz</a>

Result on the web

—

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

Employing nonlinear transformation of datasets to train neural networks

Original language description

Invited lecture at the Technical Computing Prague 2024. It is often a good idea to linearly transform each input and output signal to have a mean of zero and a standard deviation of one. This technique, called standardisation or z-scoring, is particularly useful for training statistical classifiers and recurrent neural networks because it helps with the conditionality and numerical stability of the training process. However, some problems require the neural network to predict outputs with the same relative error over different orders of magnitude. A typical class of such problems is modelling flow, vibration and other dynamic processes. This presentation shows how the training data can be nonlinearly transformed, how the transformation affects the network performance, and the drawbacks of this method. The application of nonlinear transformation of training data is also demonstrated by modelling hydrodynamic lubrication in a journal bearing using feedforward neural networks

Czech name

—

Czech description

—

Type

O - Miscellaneous

CEP classification

—

OECD FORD branch

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Project

—

Continuities

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Publication year

2024

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů