Autonomous agents must actively control the information flow surrounding them while operating in the presence of other cooperative or adversarial agents. Controlling information flow requires concealing sensitive information to succeed in adversarial domains, detaching from unnecessary information for robustness to uncertainties, and conveying information to cooperators for efficient cooperation. In this talk, I will present decision-making and learning algorithms to equip autonomous agents with such information control skills. First, I will present a deception framework where agents conceal their hostile identities from their supervisors by synthesizing and learning deceptive strategies. I will show that while deception can be done efficiently, counter-deception requires extensive computational resources. Then, I will present an analysis of model-based offline reinforcement learning that agents use in scenarios where online data collection may not be possible due to adversaries or data collection costs. Furthermore, I will consider a multiagent team planning problem under communication losses and provide performance guarantees for the team depending on the dependencies between the team members. Finally, I will present future directions on scalable autonomous deception in the presence of human and artificial intelligence vulnerabilities, integration of minimal information decision-making algorithms with learning models, and influence-aware decision-making and learning intentions in multiagent environments.

Dr. Mustafa Karabag is a postdoctoral fellow at the Oden Institute for Computational Engineering & Sciences at the University of Texas at Austin. He received his M.S. and Ph.D. degrees in electrical and computer engineering from the University of Texas at Austin in 2019 and 2023, respectively. His research draws from elements of control, reinforcement learning, optimization, and information theory to develop theory and algorithms for decision-making and learning in autonomous systems, especially for adversarial or information-scarce environments