On Hyper-Parameter Estimation in Empirical Bayes: A Revisit of the MacKay Algorithm
PRESENTER:
Yongyi Mao
University of Ottawa
ABSTRACT:
An iterative procedure introduced in MacKay's evidence framework is often
used for estimating the hyper-parameter in empirical Bayes. Despite its
effectiveness, the procedure has stayed primarily as a heuristic to date.
This paper formally investigates the mathematical nature of this procedure
and justifies it as a well-principled algorithm framework. This framework,
which we call the MacKay algorithm, is shown to be closely related to the
EM algorithm under certain Gaussian assumption.
This work was presented in UAI 2016 this summer, and is a joint work
with Chune Li, Richong Zhang and Jinpeng Huai. The PDF file of the paper
can be found at the following URL:
http://auai.org/uai2016/proceedings/papers/267.pdf
Bio:
Yongyi Mao is a Professor at the University of Ottawa. He
obtained his PhD degree at the University of Toronto in 2003, and became a
faculty member at uOttawa ever since. His research ranges over areas of
information theory and machine learning. Currently he is an Associate
Editor for IEEE Transactions on Information Theory.