667 per centimeter

a preview of things to come

—

by

ecoquant
in

I don’t post here very often, but I thought an audience might like to know what things I am working on.

I have a paper in the works regarding the “error in variables” problem, solved from a Bayesian perspective. This is where predictor variable estimates themselves have errors in their statement, as like response variables typically do.

I have a paper in the works regarding applying Professor James Spall’s SPSA to the problem of estimating the mode of a Bayesian posterior.

I also have an outline of a paper which addresses the problem of estimating a time-dependent event density given only actual reported events and a collection of reports of events at specified time stamps.

Comments

2 responses to “a preview of things to come”

  1. drewancameron Avatar
    drewancameron

    Regarding the errors-in-variables problem, there’s a highly cited paper in astronomy ( http://adsabs.harvard.edu/abs/2007ApJ…665.1489K : free version on arxiv) presenting a reasonable semi-Bayesian solution, though I find it a bit unsatisfactory. Basically, they use something like an EM algorithm to fit a Normal mixture model representing the underlying distribution of the observed predictor variable(s), and then hold this fixed for the subsequent Bayesian linear regression step. (I would prefer that the parameters of the mixture model, including the number of components, were given a fully Bayesian treatment.)

    Reply
    1. hypergeometric Avatar
      hypergeometric

      This paper is now done, although it could be polished up. I intended it as a tutorial for Bala at Duke. See https://drive.google.com/file/d/0B3Nnyie7hrIuQnQzbmJlTm9CZFk/edit?usp=sharing

      Reply

Leave a Reply

Discover more from 667 per centimeter

Subscribe now to keep reading and get access to the full archive.

Continue reading