Zinovi Rabinovich
Assistant Professor
Research Interests
Artificial Intelligence, Multi-Agent Systems, Multi-Agent Learning
Specific Research Interests
Modelling of manipulative interactions, inclusive of machine-human interactions
Perceptual control and emergent behaviour
Current Research
Current research projects focus largely on Behaviour Cultivation (a.k.a. Backdoor attacks, a.k.a. Environment Poisoning, a.k.a. Dynamic Difficulty Adjustment) against Reinforcement (Single, Multi-Agent, Distributed/Federated) Learning. Naturally, we use deep learning methods, but also rely on game theory, information theory, and evolutionary/genetic methods.
With core methodology available, we now look into:
- scale (more agents, more complex interactions and domains)
- efficiency (data needed, attack timing)
- stealth and targeting (impacting a specific portion of a multi-agent system, obfuscating attack information)
- applicability (attacking a greater variety of learning algorithms and frameworks, reducing knowledge of the targeted agent’s structures and motivations)
Biography
Dr. Rabinovich is currently a member of the School of Computer Science faculty at the Carleton University, Ottawa, where I have moved from the Nanyang Technological University, Singapore.
His prior experience also includes: the Research and Development Department of Mobileye Vision Technologies Ltd.; a Postdoctoral Researcher appointment at the Department of Computer Science at the Bar-Ilan University; and a Research Fellow position with the School of Electronics and Computer Sciences at the University of Southampton.
Ph.D. in Computer Science was conferred on me by the Hebrew University of Jerusalem in 2008.