<div dir="ltr"><div><div><br><br><br></div>Hi Folks,<br><br></div>This Friday in the YPNG seminar Jiantao Jiao (Stanford) will talk about:<br><br><div><div><div dir="ltr"><div><div>Title: Minimax estimation of the total variation distance </div><div><br>Abstract:
We consider the problem of estimating the total variation distance
<br>between two discrete distributions P and Q in a nonasymptotic and
potentially <br>large alphabet setting. When we know Q and obtain n samples
from P, we <br>show that for every Q, the minimax rate optimal estimator
with n samples <br>achieves performance comparable to the maximum likelihood
estimator with <br>n log(n) samples. When both P and Q are unknown, we
construct minimax <br>rate optimal schemes whose worst case performance is
essentially that of <br>the known Q case with Q being uniform, implying that
Q being uniform is <br>essentially the most difficult case. We discuss the
underlying techniques <br>in upper and minimax lower bounds, as well as
their potential applications <br>to other problems. </div></div></div><br>See you Friday at 11am in the Stat's classroom.<br><br>Regards,<br>sekhar</div></div></div>