Thomas Lumley • 7/26/2017

Tail bounds under sparse correlation

This in-depth statistical article discusses deriving tail bounds (probability of deviations) for sums of non-independent random variables exhibiting 'sparse correlation', a structure common in scenarios like multi-rater reliability studies (e.g., radiologists rating images). It extends ideas from Bernstein's Inequality, focusing on cases where dependencies are limited, defined via a dependency graph with parameters M (max neighborhood size) and m (size of largest independent set).

0 comments

#statistics #Probability Bounds #Bernoulli Inequality