Interdisciplinary Opportunities with AI

Abstract: Understanding human behaviour is essential for effective communication and social interaction. Gestural cues such as eye gaze, facial expressions, and body movements offer deep insights into complex mental and emotional states. Accurately capturing and modelling these cues in unconstrained environments remains a significant challenge due to data annotation complexity, variability in human behaviour, and computational limitations. My research focuses on developing robust computational models that can automatically learn and interpret non-verbal gestural cues in naturalistic, real-world scenarios, with minimal supervision. These models aim to bridge the gap between human cognitive understanding and machine perception, making AI systems more socially aware and context-sensitive.

My work has wide-ranging applications in real-world settings, including monitoring student engagement in classrooms, assessing group cohesion in collaborative environments, detecting emergent leaders, identifying dominant individuals in discussions, and analysing crowd behaviour. By advancing automatic behaviour understanding 'in-the-wild', my research contributes to building more adaptive and human-centric AI systems capable of functioning effectively in everyday social environments.

Abstract: Obscuring sensitive data from adversarial eavesdropping in distributed decision-making is a significant challenge across a range of applications, including smart grids, autonomous vehicle coordination, and multi-robot surveillance. Existing methods—such as encryption or differential privacy—modify transmitted messages without considering the underlying dynamics of the cyber-physical processes that generate the data, nor the capabilities of potential adversaries. In our recent work, we demonstrate that every dynamical system possesses an inherent system-level privacy that can be harnessed through the design of intelligent inter-agent communication protocols.

In this talk, I will present our recent results on extracting system-level privacy in the presence of eavesdropping adversaries. We derive analytical expressions that quantify the level of protection as a function of both the adversary's capabilities and the agent communication graph. Finally, I will show how these results can be applied in the context of distributed optimization and federated learning.

Speaker Bio: Dipankar Maity is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of North Carolina at Charlotte. He earned his B.E. degree in Electronics and Telecommunication Engineering from Jadavpur University, India, in 2013, and his Ph.D. degree in Electrical Engineering from the University of Maryland at College Park, in 2018. Following his doctoral studies, he was a Postdoctoral Fellow at the Georgia Institute of Technology. He has been a visiting scholar at the Technical University of Munich, Germany, and at the KTH Royal Institute of Technology, Sweden. He is a member of the IEEE Control System Society Technical Committee on Security and Privacy. He is an Associate Editor for the IEEE Control Systems Letters journal.