Abstract
This chapter provides a derivation of the Hessian of the error function E, which is required for the Bayesian evidence scheme of chapters 10 and 11. The derivation is based on an extended version of the EM algorithm, which allows the full Hessian to be decomposed into three additive components. The derivation of the first term, the Hessian of the EM error function U, is straightforward. The second term, the outer product of the gradient of the EM error function, is found to be cancelled out. An approximation is made for the third term, the expectation value for the outer product of the gradient of Ψ, which is approximated by a diagonal block matrix. The justification for this simplification is given in the text.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
Reference
This approximation is not always good, but recall from chapters 10 and 11 that the off-diagonal terms Hαi,wk and Hβiwk are not needed for the evidence scheme
For alternatives schemes, that make use of the whole Hessian (e.g. [26]), approximation (18.45) can be applied.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 1999 Springer-Verlag London Limited
About this chapter
Cite this chapter
Husmeier, D. (1999). Appendix: Derivation of the Hessian for the Bayesian Evidence Scheme. In: Neural Networks for Conditional Probability Estimation. Perspectives in Neural Computing. Springer, London. https://doi.org/10.1007/978-1-4471-0847-4_18
Download citation
DOI: https://doi.org/10.1007/978-1-4471-0847-4_18
Publisher Name: Springer, London
Print ISBN: 978-1-85233-095-8
Online ISBN: 978-1-4471-0847-4
eBook Packages: Springer Book Archive