Home // INTELLI 2015, The Fourth International Conference on Intelligent Systems and Applications // View article

Comparing Knowledge Representation Forms in Empirical Model Building

Authors:
Hao Wang
Ingo Schwab
Michael Emmerich

Keywords: Function Approximation; Machine learning; Process Modelling; Model Formation; Symbolic Regression

Abstract:
Empirical models in engineering practice often come from measurements of the machines but might also be generated from expensive simulations to build so-called surrogate models. From an abstract point of view can be seen as approximations of functions that map input variables to output variables. This paper describes and conceptually compares different function approximation techniques, with a focus on methods from machine learning, including Kriging Models, Gaussian Processes, Artificial Neural Networks, Radial Basis Functions, Random Forests, Functional Regression, and Symbolic Regression. These methods are compared on basis of different criteria, such as speed, number and type of parameters, uncertainty assessment, interpretability, and smoothness properties. Besides, a particular focus is to compare the different ways of how knowledge is represented in these models. Here we compare the families of functions used to build the model and which model components (structures, parameters) are provided by the user or learned from the available data. Although this paper is not about benchmarking, some numerical examples are provided that illustrate the typical behavior of the methods.

Pages: 170 to 178

Copyright: Copyright (c) IARIA, 2015

Publication date: October 11, 2015

Published in: conference

ISSN: 2308-4065

ISBN: 978-1-61208-437-4

Location: St. Julians, Malta

Dates: from October 11, 2015 to October 16, 2015