Home > Research > Publications & Outputs > Comment on "Inference with minimal Gibbs free e...

Associated organisational unit

Electronic data

Links

Text available via DOI:

View graph of relations

Comment on "Inference with minimal Gibbs free energy in information field theory"

Research output: Contribution to Journal/MagazineEditorialpeer-review

Published
Article number033101
<mark>Journal publication date</mark>20/03/2012
<mark>Journal</mark>Physical Review E
Issue number3
Volume85
Number of pages4
Pages (from-to)-
Publication StatusPublished
<mark>Original language</mark>English

Abstract

Ensslin and Weig [Phys. Rev. E 82, 051112 (2010)] have introduced a "minimum Gibbs free energy" (MGFE) approach for estimation of the mean signal and signal uncertainty in Bayesian inference problems: it aims to combine the maximum a posteriori (MAP) and maximum entropy (ME) principles. We point out, however, that there are some important questions to be clarified before the new approach can be considered fully justified, and therefore able to be used with confidence. In particular, after obtaining a Gaussian approximation to the posterior in terms of the MGFE at some temperature T, this approximation should always be raised to the power of T to yield a reliable estimate. In addition, we show explicitly that MGFE indeed incorporates the MAP principle, as well as the MDI (minimum discrimination information) approach, but not the well-known ME principle of Jaynes [E.T. Jaynes, Phys. Rev. 106, 620 (1957)]. We also illuminate some related issues and resolve apparent discrepancies. Finally, we investigate the performance of MGFE estimation for different values of T, and we discuss the advantages and shortcomings of the approach.

Bibliographic note

©2012 American Physical Society