BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Adaptive Data Analysis and Error-Control via Information Measures
DTSTART:20180821T140000
DTEND:20180821T160000
DTSTAMP:20260406T212816Z
UID:52d9f774f8b218f95d75540c168459b1c172fe86a4f29cf325f60bd4
CATEGORIES:Conferences - Seminars
DESCRIPTION:Amedeo Esposito\nEDIC candidacy exam\nExam president: Prof. Ol
 ivier Lévêque\nThesis advisor: Prof. Michael Gastpar\nCo-examiner: Prof.
  Volkan Cevher\n\nAbstract\nThere is an increasing concern that most curre
 nt published research findings are false. The main cause seems to lie in t
 he fundamental disconnection between theory and practise in data analysis.
  While the former typically relies on statistical independence\, the latte
 r is an inherently adaptive process: new hypotheses are formulated based o
 n the outcomes of previous analyses. A recent line of work tries to mitiga
 te these issues using mechanisms\, like Differential Privacy\, that compos
 e adaptively while degrading gracefully and thus provide statistical guara
 ntees even in adaptive contexts. Our contribution consists in the introduc
 tion of a new approach\, based on the concept of Maximal Leakage\, an info
 rmation-theoretic measure of leakage of information. We will see how this 
 measure ensures generalization guarantees\, composes adaptively and is rob
 ust under post-processing\, making it a good candidate for being used in t
 his framework.\n\nBackground papers\nAn operational measure of Information
  Leakage\, by Issa\, I.\, et al.\nGeneralization in Adaptive Data Analysis
  and Holdout Reuse by Cynthia Dwork et al.\nMax-Information\, Differential
  Privacy and Post-selection Hypothesis Testing\, Rogers\, R.\, et al.\n \
 n 
LOCATION:BC 129 https://plan.epfl.ch/?room==BC%20129
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR