Covariance-based weighting for optimal combination of network predictions

Conference item

This paper introduces a method for calculating the covariance between different neural network solutions. It is based on a generalisation of the delta method for calculating the network Hessian and generates what we call the 'cross-covariance' matrix (its inverse is the 'cross-Hessian'). Using this matrix we are able to estimate the covariance between network predictions at each point in input space, using training data alone. Whilst this is a significant result in itself we have also applied...

Expand abstract

Authors: Penny, WD; Husmeier, D; Roberts, SJ

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Institution of Engineering and Technology Publisher's website
• Host title: 1999 Ninth International Conference on Artificial Neural Networks ICANN 99. (Conf. Publ. No. 470)
• Volume: 2
• Pages: 826-831
• Publication date: 1999-12-31
• Event title: Ninth International Conference on Artificial Neural Networks (ICANN 99)
• Event location: Edinburgh, UK
• Event start date: 1999-09-07
• Event end date: 1999-09-10
• DOI: 10.1049/cp:19991214
• ISSN: 0537-9989
• ISBN: 0852967217
• Language: English
• Keywords: SBTMR; model predictions; neural network; covariance matrix; generalisation; Taylor series; learning; covariance-based weighting