BEGIN:VCALENDAR VERSION:2.0 PRODID:-//wp-events-plugin.com//7.4.0.1//EN TZID:Europe/Paris X-WR-TIMEZONE:Europe/Paris BEGIN:VEVENT UID:148@isdm.umontpellier.fr DTSTART;TZID=Europe/Paris:20250404T110000 DTEND;TZID=Europe/Paris:20250404T110000 DTSTAMP:20260602T130233Z URL:https://isdm.umontpellier.fr/events/optimal-classification-under-perfo rmative-distribution-shift-2/ SUMMARY:Optimal Classification under Performative Distribution Shift DESCRIPTION:Campus St Priest (860 Rue Saint Priest 34095 Montpellier Cedex 5)\, bat. 5\, Room: 02.124\nMachine Learning in Montpellier\, Theory &\ ; Practice - Olivier Cappé (CNRS / ENS / Université PSL)\n\nPerformative learning addresses the increasingly pervasive situations in which algorit hmic decisions may induce changes in the data distribution as a consequenc e of their public deployment. We propose a novel view in which these perfo rmative effects are modelled as push-forward measures. This general framew ork encompasses existing models and enables novel performative gradient es timation methods\, leading to more efficient and scalable learning strateg ies. For distribution shifts\, unlike previous models which require full s pecification of the data distribution\, we only assume knowledge of the sh ift operator that represents the performative changes. Focusing on classif ication with a linear-in-parameters performative effect\, we prove the con vexity of the performative risk under a new set of assumptions. We also es tablish a connection with adversarially robust classification by reformula ting the minimization of the performative risk as a min-max variational pr oblem.\n\nIA&\;Experts ATTACH;FMTTYPE=image/jpeg:https://isdm.umontpellier.fr/wp-content/uploads/ 2025/02/ml-mpt-1.jpg END:VEVENT BEGIN:VTIMEZONE TZID:Europe/Paris X-LIC-LOCATION:Europe/Paris BEGIN:DAYLIGHT DTSTART:20250330T030000 TZOFFSETFROM:+0100 TZOFFSETTO:+0200 TZNAME:CEST END:DAYLIGHT END:VTIMEZONE END:VCALENDAR