BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:IACR AI/ML Seminar: Busting the Paper Ballot: Voting Meets Adv
 ersarial Machine Learning
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20261007T193754Z
UID:tag:localist.com\,2008:EventInstance_51160950328713
DTSTART:20251204T200000Z
DTEND:20251204T213000Z
DESCRIPTION:Please see below for the next talk in the fall seminar series o
 rganized by the Institute for AI & Computational Research on AI/ML techniq
 ues and applications across various scientific domains. You can find a tab
 le of upcoming talks here: https://web.uri.edu/iacr/seminars/.\n\n \n\nSpe
 aker: Kaleel Mahmood  (URI)\n\nDate/Time/Location: Dec 4\, 3pm\, Memorial 
 Union room 308.\n\nTitle: Busting the Paper Ballot: Voting Meets Adversari
 al Machine Learning\n\n\nAbstract: Can adversarial machine learning change
  the outcome of US election? We present the security risk associated with 
 using machine learning classifiers in United States election tabulators. T
 he central classification task in election tabulation is deciding whether 
 a mark does or does not appear on a bubble associated to an alternative in
  a contest on the ballot. Barretto et al. (E-Vote-ID 2021) reported that c
 onvolutional neural networks are a viable option in this field\, as they o
 utperform simple feature-based classifiers. Our contributions to election 
 security can be divided into four parts. To demonstrate and analyze the hy
 pothetical vulnerability of machine learning models on election tabulators
 \, we first introduce four new ballot datasets. Second\, we train and test
  a variety of different models on our new datasets. These models include s
 upport vector machines\, convolutional neural networks (a basic CNN\, VGG 
 and ResNet)\, and vision transformers (Twins and CaiT). Third\, using our 
 new datasets and trained models\, we demonstrate that traditional white bo
 x attacks are ineffective in the voting domain due to gradient masking. Ou
 r analyses further reveal that gradient masking is a product of numerical 
 instability. We use a modified difference of logits ratio loss to overcome
  this issue (Croce and Hein\, ICML 2020). Fourth\, in the physical world\,
  we conduct attacks with the adversarial examples generated using our new 
 methods. In traditional adversarial machine learning\, a high (50% or grea
 ter) attack success rate is ideal. However\, for certain elections\, even 
 a 5% attack success rate can flip the outcome of a race. We show such an i
 mpact is possible in the physical domain.
GEO:41.485031;-71.529778
LOCATION:Memorial Union\, 308
SUMMARY:IACR AI/ML Seminar: Busting the Paper Ballot: Voting Meets Adversar
 ial Machine Learning
URL;VALUE=URI:https://events.uri.edu/event/iacr-aiml-seminar-busting-the-pa
 per-ballot-voting-meets-adversarial-machine-learning
CATEGORIES:Lectures / Presentations
END:VEVENT
END:VCALENDAR