BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Characterizing Machine Unlearning through Definitions and Implemen tations - Nicolas Papernot\, University of Toronto and Vector Institute DTSTART:20240229T140000Z DTEND:20240229T150000Z UID:TALK210715@talks.cam.ac.uk CONTACT:Hridoy Sankar Dutta DESCRIPTION:The talk presents open problems in the study of machine unlear ning. The need for machine unlearning\, i.e.\, obtaining a model one would get without training on a subset of data\, arises from privacy legislatio n and as a potential solution to data poisoning or copyright claims. The f irst part of the talk discusses approaches that provide exact unlearning: these approaches output the same distribution of models as would have been obtained by training without the subset of data to be unlearned in the fi rst place. While such approaches can be computationally expensive\, we dis cuss why it is difficult to relax the guarantee they provide to pave the w ay for more efficient approaches. The second part of the talk asks if we c an verify unlearning. Here we show how an entity can claim plausible denia bility when challenged about an unlearning request that was claimed to be processed\, and conclude that at the level of model weights\, being unlear nt is not always a well-defined property. Instead\, unlearning is an algor ithmic property.\n\nRECORDING : Please note\, this event will be recorded and will be available after the event for an indeterminate period under a CC BY -NC-ND license. Audience members should bear this in mind before joi ning the webinar or asking questions.\n\nhttps://cam-ac-uk.zoom.us/j/82112 795708?pwd=VFBTVjI1YkRqMXY5MEpRcXYzdmN6QT09\n\nMeeting ID: 821 1279 5708\n Passcode: 468381\n\nNOTE: Please do not post URLs for the talk\, and espec ially Zoom links to Twitter because automated systems will pick them up an d disrupt our meeting. LOCATION:Webinar &\; FW11\, Computer Laboratory\, William Gates Buildin g. END:VEVENT END:VCALENDAR