skip to content

Strong converses and high-dimensional statistical estimation problems

Presented by: 
Ramji Venkataramanan
Tuesday 24th July 2018 - 09:45 to 10:30
INI Seminar Room 1
In many statistical inference problems, we wish to bound the performance of any possible estimator. This can be seen as a converse result, in a standard information-theoretic sense. A standard approach in the statistical literature is based on Fano’s inequality, which typically gives a weak converse. We adapt these arguments by replacing Fano by more recent information-theoretic ideas, based on the work of Polyanskiy, Poor and Verdu. This gives tighter lower bounds that can be easily computed and are asymptotically sharp. We illustrate our technique in three applications: density estimation, active learning of a binary classifier, and compressed sensing, obtaining tighter risk lower bounds in each case.  

(joint with Oliver Johnson, see doi:10.1214/18-EJS14)

The video for this talk should appear here if JavaScript is enabled.
If it doesn't, something may have gone wrong with our embedded player.
We'll get it fixed as soon as possible.
University of Cambridge Research Councils UK
    Clay Mathematics Institute London Mathematical Society NM Rothschild and Sons