Current Institution: The University of Texas at Austin
Bio: I am currently a fourth-year Ph.D. student in the Department of Electrical and Computer Engineering at the University of Texas at Austin and a member of Professor Ufuk Topcu’s Autonomous Systems Group. Before my graduate studies in Austin, I received my Bachelor’s and Master’s degrees in Aerospace Engineering from ISAE-SUPAERO, France, in 2018. I also received a Master’s degree in Computer Science from École Polytechnique, France, in 2017.
My research interests lie in enabling artificial intelligence-powered autonomous systems to learn and reason like humans do, i.e., safely learn complex and uncorrelated skills through incredibly small amounts of interactions with the world around them. To endow autonomous systems with such capabilities, I develop theoretical, practical, and computational methods by drawing from diverse fields such as control theory, formal methods, data-driven modeling and control, reinforcement learning, and optimization.
Abstract: Toward common sense in learning for autonomous systems
Humans can learn complex skills through incredibly small amounts of interactions with the world around them. Empirical evidence suggests that effective utilization of common sense plays a crucial role in human’s effective learning. In contrast, modern learning techniques rely on excessive amounts of data and often fail to generalize to unseen environments.
My research focuses on the following question: What can play the role of common sense in artificial intelligence (AI)? I argue that humans’ vast knowledge of how the world works can simulate the effect of common sense into AI. Such knowledge may be known underlying rules in the environment, contextual information such as task specifications, or structural knowledge such as physics-driven constraints. By leveraging prior knowledge into learning, I have developed algorithms that offer unprecedented data efficiency and generalization beyond the training regime. My research focuses on practical and computationally eﬀicient algorithms with provable performance and safety guarantees.