BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Machine Learning in context of Computer Security - Ilia Shumailov\
 , University of Cambridge
DTSTART:20220215T140000Z
DTEND:20220215T150000Z
UID:TALK170492@talks.cam.ac.uk
CONTACT:Kieron Ivy Turk
DESCRIPTION:Machine learning (ML) has proven to be more fragile than previ
 ously thought\, especially in adversarial settings. A capable adversary ca
 n cause ML systems to break at training\, inference\, and deployment stage
 s. In this talk\, I will cover my recent work on attacking and defending m
 achine learning pipelines\; I will describe how\, otherwise correct\, ML c
 omponents end up being vulnerable because an attacker can break their unde
 rlying assumptions. First\, with an example of attacks against text prepro
 cessing\, I will discuss why a holistic view of the ML deployment is a key
  requirement for ML security. Second\, I will describe how an adversary ca
 n exploit the computer systems\, underlying the ML pipeline\, to develop a
 vailability attacks at both training and inference stages. At the training
  stage\, I will present data ordering attacks that break stochastic optimi
 sation routines. At the inference stage\, I will describe sponge examples 
 that soak up a large amount of energy and take a long time to process. Fin
 ally\, building on my experience attacking ML systems\, I will discuss dev
 eloping robust defenses against ML attacks\, which consider an end-to-end 
 view of the ML pipeline.
LOCATION:Webinar &amp\; LT2\, Computer Laboratory\, William Gates Building
 .
END:VEVENT
END:VCALENDAR