Title
Evading AI systems
Abstract
Machine Learning based systems are being deployed in more and more critical areas. However, they can be vulnerable to simple and difficult-to-detect attacks which allow an attacker to alter the behavior of the system to their own advantage. In this talk we will discuss two such attacks. The first, LowKey, can reliably allow individuals to avoid recognition by black-box industrial and governmental facial recognition systems. The second, Witches' Brew, allows an attacker to imperceptibly alter training data so that any system trained on that data will be generally accurate, except on individual testing examples of the attacker's choosing. This research is important for those who wish to evade AI systems, those who want to build and deploy more robust AI systems, and those who are interested in understanding the black-box nature of modern AI.
Bio
Gavin Taylor is a Professor of Computer Science at the US Naval Academy. His research spans a variety of topics in Machine Learning, including AI evasion, distributed optimization, reinforcement learning, and the fundamental behaviors of neural networks.
Practical
The Tuesday seminar series explores a wide range of topics relevant to our research themes. Each week, invited researchers present for 40 minutes, followed by a group discussion.
Seminars are open for attendance for everybody.
For those unable to attend in person: https://uio.zoom.us/j/62535744013
Speaker is presenting online.
Point of contacts:
- Organiser and moderator: Centre Director Arnoldo Frigessi, University of Oslo
- Integreat Administrative Leader Maria Dikova, University of Oslo