skip to content

Function estimation on large graphs with missing data

Presented by: 
Alisa Kirichenko
Monday 25th July 2016 - 11:30 to 12:00
INI Seminar Room 1
Co-author: Harry van Zanten (University of Amsterdam)

There are various problems in statistics and machine learning that involve making an inference about a function on a graph. I will present a Bayesian approach to estimating a smooth function in the context of regression and classification problems on graphs. I will discuss the mathematical framework that allows to study the performance of nonparametric function estimation methods on large graphs. I will review theoretical results that show how to achieve asymptotically optimal Bayesian regularization under geometry conditions on the families of the graphs and the smoothness assumption on the true function. Both assumptions are formulated in terms of graph Laplacian. I will also discuss the case of "uniformly distributed" missing observations and investigate the generalization performance for various missing mechanisms.
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