On the Importance of α Marginalization in Maximum Entropy

Research output: Chapter in Book/Report/Conference proceedingOther chapter contributionResearch

Abstract

The correct entropic prior, computed by marginalization over the reg- ularization parameter a, is used to invert photoemission data and to restore the famous “Susie” image. Comparison with the conventional maximum entropy procedure shows less overfitting of noise and demonstrates the residual ringing which is intrinsic to ill-posed inversion problems. An improvement to the steepest descent approximation reveals the reason for the overfitting. On top of that, the correct treatment of the regularization parameter is vital for the existence of the continuum-limit of MaxEnt.
Original languageEnglish
Title of host publicationMaximum Entropy and Bayesian Methods
EditorsKenneth M. Hanson, Richard N. Silver
PublisherSpringer Netherlands
Pages229-236
Number of pages8
ISBN (Print)978-94-010-6284-8 978-94-011-5430-7
Publication statusPublished - 1996

Publication series

NameFundamental Theories of Physics
PublisherSpringer Netherlands

Fingerprint

marginalization
entropy
ringing
parameter
comparison
inversion

Keywords

  • Electrical Engineering, Entropic Prior, Image Processing, Inverse Problem, Physics, general, Probability Theory and Stochastic Processes, Regularization, Statistical Physics, Dynamical Systems and Complexity, Statistics for Engineering, Physics, Computer Science, Chemistry and Earth Sciences, Statistics, general

Cite this

Fischer, R., Linden, W. V. D., & Dose, V. (1996). On the Importance of α Marginalization in Maximum Entropy. In K. M. Hanson, & R. N. Silver (Eds.), Maximum Entropy and Bayesian Methods (pp. 229-236). (Fundamental Theories of Physics). Springer Netherlands.

On the Importance of α Marginalization in Maximum Entropy. / Fischer, R.; Linden, W. Von Der; Dose, V.

Maximum Entropy and Bayesian Methods. ed. / Kenneth M. Hanson; Richard N. Silver. Springer Netherlands, 1996. p. 229-236 (Fundamental Theories of Physics).

Research output: Chapter in Book/Report/Conference proceedingOther chapter contributionResearch

Fischer, R, Linden, WVD & Dose, V 1996, On the Importance of α Marginalization in Maximum Entropy. in KM Hanson & RN Silver (eds), Maximum Entropy and Bayesian Methods. Fundamental Theories of Physics, Springer Netherlands, pp. 229-236.
Fischer R, Linden WVD, Dose V. On the Importance of α Marginalization in Maximum Entropy. In Hanson KM, Silver RN, editors, Maximum Entropy and Bayesian Methods. Springer Netherlands. 1996. p. 229-236. (Fundamental Theories of Physics).
Fischer, R. ; Linden, W. Von Der ; Dose, V. / On the Importance of α Marginalization in Maximum Entropy. Maximum Entropy and Bayesian Methods. editor / Kenneth M. Hanson ; Richard N. Silver. Springer Netherlands, 1996. pp. 229-236 (Fundamental Theories of Physics).
@inbook{d60d301f15e04e99bc200c90e67d99c7,
title = "On the Importance of α Marginalization in Maximum Entropy",
abstract = "The correct entropic prior, computed by marginalization over the reg- ularization parameter a, is used to invert photoemission data and to restore the famous “Susie” image. Comparison with the conventional maximum entropy procedure shows less overfitting of noise and demonstrates the residual ringing which is intrinsic to ill-posed inversion problems. An improvement to the steepest descent approximation reveals the reason for the overfitting. On top of that, the correct treatment of the regularization parameter is vital for the existence of the continuum-limit of MaxEnt.",
keywords = "Electrical Engineering, Entropic Prior, Image Processing, Inverse Problem, Physics, general, Probability Theory and Stochastic Processes, Regularization, Statistical Physics, Dynamical Systems and Complexity, Statistics for Engineering, Physics, Computer Science, Chemistry and Earth Sciences, Statistics, general",
author = "R. Fischer and Linden, {W. Von Der} and V. Dose",
note = "DOI: 10.1007/978-94-011-5430-727",
year = "1996",
language = "English",
isbn = "978-94-010-6284-8 978-94-011-5430-7",
series = "Fundamental Theories of Physics",
publisher = "Springer Netherlands",
pages = "229--236",
editor = "Hanson, {Kenneth M.} and Silver, {Richard N.}",
booktitle = "Maximum Entropy and Bayesian Methods",
address = "Netherlands",

}

TY - CHAP

T1 - On the Importance of α Marginalization in Maximum Entropy

AU - Fischer, R.

AU - Linden, W. Von Der

AU - Dose, V.

N1 - DOI: 10.1007/978-94-011-5430-727

PY - 1996

Y1 - 1996

N2 - The correct entropic prior, computed by marginalization over the reg- ularization parameter a, is used to invert photoemission data and to restore the famous “Susie” image. Comparison with the conventional maximum entropy procedure shows less overfitting of noise and demonstrates the residual ringing which is intrinsic to ill-posed inversion problems. An improvement to the steepest descent approximation reveals the reason for the overfitting. On top of that, the correct treatment of the regularization parameter is vital for the existence of the continuum-limit of MaxEnt.

AB - The correct entropic prior, computed by marginalization over the reg- ularization parameter a, is used to invert photoemission data and to restore the famous “Susie” image. Comparison with the conventional maximum entropy procedure shows less overfitting of noise and demonstrates the residual ringing which is intrinsic to ill-posed inversion problems. An improvement to the steepest descent approximation reveals the reason for the overfitting. On top of that, the correct treatment of the regularization parameter is vital for the existence of the continuum-limit of MaxEnt.

KW - Electrical Engineering, Entropic Prior, Image Processing, Inverse Problem, Physics, general, Probability Theory and Stochastic Processes, Regularization, Statistical Physics, Dynamical Systems and Complexity, Statistics for Engineering, Physics, Computer

M3 - Other chapter contribution

SN - 978-94-010-6284-8 978-94-011-5430-7

T3 - Fundamental Theories of Physics

SP - 229

EP - 236

BT - Maximum Entropy and Bayesian Methods

A2 - Hanson, Kenneth M.

A2 - Silver, Richard N.

PB - Springer Netherlands

ER -