BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:"Statistical graph clustering"
DTSTART:20170126T100000
DTEND:20170126T110000
DTSTAMP:20260407T145914Z
UID:949aa932b25275b0c923876954ecc52142529bc18b0abe3e33f26780
CATEGORIES:Conferences - Seminars
DESCRIPTION:Prof. Emmanuel Abbe (Princeton University)\nClustering is at t
 he core of unsupervised machine learning and network data analysis. Despit
 e significant advances at the methodological level\, establishing fundamen
 tal limits remains a challenge. Our approach is based on a statistical fra
 mework to graph clustering (community detection)\, starting with stochasti
 c block models. In this context\, we obtain sharp characterizations of whe
 n clusters estimations are possible or not. These are obtained from two pr
 oof techniques: local to global hypothesis testing and inference on branch
 ing processes. We further establish the physicists’ conjecture on the em
 ergence of a gap between the statistical and computational thresholds\, us
 ing a sampling method. This underlines the importance of bridging statisti
 cal and computational methods in this field\, likely to be a recurrent the
 me in data sciences. Data implementations\, spectral methods and connectio
 ns to Ramanujan graphs will also be discussed.
LOCATION:BI A0 448 https://plan.epfl.ch/?room==BI%20A0%20448
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR