Speaker: Mr. Chuanhao Li
From: University of Virginia
Abstract
Interactive decision making (e.g., bandit and reinforcement learning) is a promising paradigm for developing intelligent systems capable of efficiently exploring unknown environments. It has achieved remarkable success in recommender systems, clinical decision systems, robotics and cyber-physical systems, game playing, etc. Most prior efforts in the theoretical foundations of this paradigm consider the single agent setting, with strategic exploration being the primary focus. However, in real-world scenarios, intelligent systems typically involve multiple agents, e.g., AI and human beings, which gives rise to both new challenges and new opportunities.
In this talk, I will first introduce our recent works that enable collaborative decision making under various practical scenarios, e.g., heterogeneous, non-stationary, and distributed environments, with improved sample efficiency guarantee by utilizing agents’ task similarities. I will then present our works that further consider interactive decision making with strategic agents, and discuss, as the system designer, how to cope with different strategic behaviors of the agents to achieve desirable decision outcomes. I will also share potential future directions of interactive decision making and visions to build practical decision-making systems for real-world applications.
For more info, please follow this link.
Read More