A Comparative Study of Different Techniques for Improving the Robustness of Neural Network Models (2000)

Student name: ABD HALIM SHAH MAULUD

Degree and date obtained: MSc in Applied Process Control (Distinction), October 2000

Career after study: University Lecturer in Malaysia

Thesis Abstract: Artificial Neural Networks have been shown to be able to approximate any continuous non-linear functions and have been used to build data-based empirical models for non-linear processes. The main advantage of neural network based process models is that they are easy to build. This feature is particularly useful when modelling complicated processes where detailed mechanistic models are difficult to build. However, a critical shortcoming of neural network models is that they often lack robustness unless a proper network training and validation procedure is used. Here model robustness means that a model can still give satisfactory predictions when applied to new (unseen) data. Model robustness is a basic requirement in advanced process monitoring and control. Several techniques have been recently developed to improve neural network model robustness, such as training with regularisation, cross validation based "early stopping", stacked neural networks, committees of networks, and bootstrap aggregated neural networks. This project performs a comparative study of some of these techniques and investigates how these techniques can be combined. Some benchmark problems are used as case studies.