Talks
Fall 2016

# Monotone Estimation Framework and its Applications for Scalable Analytics of Large Data Sets

Friday, Nov. 18, 2016 11:20 am12:00 pm PST

Monotone estimation problems (MEP) have a simple formulation: We have a data point $x$ from a universe $D$ and we are interested in estimating $f(x)$ for a nonnegative function $f$. However, we do not observe $x$ but instead see the output $S(x,u)$ of a blurry scope applied to $x$ with resolution $u \sim U[0,1]$: The lower $u$ is, the more information we have on $x$ and hence on $f(x)$. We are interested in estimators that are unbiased, nonnegative, and optimal (admissible).