A key issue in the use of trainable systems is to test the reliability achievable in real-world applications. In this context it is important to assess and quantify the effects that the inclusion of these methods may have on systems in which they are embedded.
The aim of the research project Validation and Verification of Neural Network Systems is to provide a set of principled methods for addressing the requirement of systems which embed neural networks. This project which is funded at the Neural Computing Research Group of Aston University is jointly supported by Lloyd's Register, AEA Technology, British Aerospace and by the EPSRC under the Neural Network key questions programme.
The component of the 3-year project which involves the
Neural Computing Research Group in collaboration with
British Aerospace Sowerby Research Centre looks at the
theoretical principles behind neural networks and how the practical
implementation of these can be used to assess their dependability.
As an aid to this study the problem of classification of segmented images
has been used.
Project Home page | Project Overview | Work plan | Publications |