20/01/2020
A special talk on "Security, Privacy and Machine Learning" by Prof. N. Asokan, Computer Science at the University of Waterloo.
Title: Security, Privacy and Machine Learning
Speaker: Professor N. Asokan, Computer Science at the University of Waterloo
Date: Friday 24th January 2020
Time: 1:00-2:00 pm.
Place: Room 106, CSIM Building
Abstract:
The dramatic successes in applying machine learning (ML) to a wide variety of areas have resulted in a huge spike in the popularity of ML. But success invites the attention of miscreants who seek to compromise successful systems for fun and profit. In this talk, I will argue that taking a systems security approach to ML-based systems will allow us to systematically identify various security/privacy concerns in those systems. I will briefly outline several such examples, and discuss one, model extraction, in more detail. Model extraction allows a malicious client of an ML-based system to "steal" the functionality of the underlying ML model by only using its access to the prediction API. I will describe our recent experiments which show that with a strong, but realistic, adversary, model extraction attacks are not only feasible but also hard to detect or prevent. I will briefly discuss watermarking as a possible deterrent against model extraction.
Speaker bio:
N. Asokan is a Professor of Computer Science at the University of Waterloo, where he holds a David R. Cheriton Chair. He is also an Adjunct Professor of Computer Science at Aalto University. Prior to joining academia, Asokan spent several years in industrial research, first at IBM Research and then at Nokia Research Center. Asokan is broadly interested in systems security, with particular emphases on (mobile) platform security, and the interplay between security/privacy and machine learning. Asokan is a Fellow of the IEEE and the ACM, and was awarded the ACM SIGSAC Outstanding Innovation Award in 2018. For more information about Asokan and his research visit his website (https://asokan.org/asokan/) or follow him on Twitter ().
We are looking forward to seeing you in the event.
