Seminario
"Bias and Variance Asymptotic expansions of Maximum likelihood Estimators(MLE): theoretical developments and applications."
Relatore: Michele Zanolin MIT  Boston
Aula Einstein  Dipartimento di Fisica
02 Ottobre 2003 ore 16.30
Abstract

Bias and Variance Asymptotic expansions of Maximum
likelihood Estimators(MLE): theoretical developments
and applications.
Parameter estimations from measurements of Gaussian
observables that are non linearly related with the
parameters,or in general from non Gaussian observables,
typically presents prohibitive difficulties in
quantifyingthe estimate's error. Furthermore, even if the
estimate is unbiased, the variance often exceeds the
theoretical minimum (CramerRao lower bound)by orders of
magnitude.
The problem is discussed here by means of asymptotic
expansions of MLE's bias and variances in inverse
powers of the number of statistically independent samples of
the observable.It is also discussed how these expansions,
for Gaussian data, can be reexpressed in terms of inverse
powers of the signal to noise ratio.
Applications to Ocean waveguides, to parameter estimation
from gravitational waves, and if there is enough time, to
the search for an ocean on a moon of Jupiter will be
discussed.


