BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//https://caida.ubc.ca//NONSGML iCalcreator 2.41.92//
CALSCALE:GREGORIAN
METHOD:PUBLISH
UID:63303338-3561-4537-a533-306538643535
X-WR-RELCALID:efc09d74-9c93-479e-a94f-485231ddccde
X-WR-TIMEZONE:America/Vancouver
X-WR-CALNAME:Reconstructing Training Data from Model Gradient\, Provably - 
 Qi Lei\, Assistant Professor\, NYU
BEGIN:VTIMEZONE
TZID:America/Vancouver
TZUNTIL:20250309T100000Z
BEGIN:STANDARD
TZNAME:PST
DTSTART:20221106T020000
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
RDATE:20231105T020000
RDATE:20241103T020000
END:STANDARD
BEGIN:DAYLIGHT
TZNAME:PDT
DTSTART:20230312T020000
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
RDATE:20240310T020000
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:665ed9e2-0995-44da-9ddc-6ff3835e32a5
DTSTAMP:20260429T105701Z
CLASS:PUBLIC
CREATED:20230704T165925Z
DESCRIPTION:Abstract: Understanding when and how much a model gradient leak
 s information about the training sample is an important question in privac
 y. In this talk\, we present a surprising result: even without training or
  memorizing the data\, we can fully reconstruct the training samples from 
 a single gradient query at a randomly chosen parameter value. We prove the
  identifiability of the training data under mild conditions: with shallow 
 or deep neural networks and a wide range of activation functions. We also 
 present a statistically and computationally efficient algorithm based on l
 ow-rank tensor…
DTSTART;TZID=America/Vancouver:20230711T113000
DTEND;TZID=America/Vancouver:20230711T123000
LAST-MODIFIED:20230705T222523Z
LOCATION:UBC Vancouver Campus\, ICCS X836 / Please register to receive Zoom
  link
SUMMARY:Reconstructing Training Data from Model Gradient\, Provably - Qi Le
 i\, Assistant Professor\, NYU
TRANSP:OPAQUE
URL:https://caida.ubc.ca/event/reconstructing-training-data-model-gradient-
 provably-qi-lei-assistant-professor-nyu
END:VEVENT
END:VCALENDAR