BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260402T024533Z
LOCATION:3006\, Level 3
DTSTART;TZID=America/Los_Angeles:20250624T104500
DTEND;TZID=America/Los_Angeles:20250624T110000
UID:dac_DAC 2025_sess148_RESEARCH335@linklings.com
SUMMARY:ReVeil: Unconstrained Concealed Backdoor Attack on Deep Neural Net
 works using Machine Unlearning
DESCRIPTION:Manaar Alam, Hithem Lamri, and Michail Maniatakos (New York Un
 iversity, Abu Dhabi)\n\nBackdoor attacks embed hidden functionalities in d
 eep neural networks (DNN), triggering malicious behavior with specific inp
 uts. Advanced defenses monitor anomalous DNN inferences to detect such att
 acks. However, concealed backdoors evade detection by maintaining a low pr
 e-deployment attack success rate (ASR) and restoring high ASR post-deploym
 ent via machine unlearning. Existing concealed backdoors are often constra
 ined by requiring white-box or black-box access or auxiliary data, limitin
 g their practicality when such access or data is unavailable. This paper i
 ntroduces ReVeil, a concealed backdoor attack targeting the data collectio
 n phase of the DNN training pipeline, requiring no model access or auxilia
 ry data. ReVeil maintains low pre-deployment ASR across four datasets and 
 four trigger patterns, successfully evades three popular backdoor detectio
 n methods, and restores high ASR post-deployment through machine unlearnin
 g.\n\nTopics: Security\n\nTracks: SEC1: AI/ML Security/Privacy\n\nSession 
 Chairs: Adnan Siraj Rakin (Binghamton University) and Ayesha Siddique (Uni
 versity of Maine)\n\n
END:VEVENT
END:VCALENDAR