Yang Cai - Research Focus | Center for Algorithms, Data, and Market Design at Yale

Multi-Agent Learning: Over the past decade, Machine Learning (ML) has experienced remarkable advancements in a multitude of applications. These accomplishments can be largely ascribed to the prevailing approach of training ML systems through the minimization of a singular loss function, facilitated by efficient optimization techniques. Nonetheless, a notable shift can be observed in contemporary ML applications, veering away from the single objective perspective. Instead, they can be more aptly conceptualized as games engaging multiple intelligent agents or algorithms. These games can be explicit, as seen in markets, traffic routing, game-solving systems such as AlphaZero, and multi-agent Reinforcement Learning systems, or implicit, as in the case of generative adversarial networks, adversarial examples, robust optimization, and so on. Game theory provides a framework for understanding such interactions between agents, each with their own utility function to optimize. Modern Machine Learning (ML) applications, however, introduce new complexities. Specifically, these associated games tend to be high-dimensional and non-concave, meaning there's a vast array of potential strategies, the agents’ utilities could depend on their strategies in complicated and non-concave ways. The main question we would like to investigate is the following: What are the meaningful solutions in non-concave high-dimensional games, and when can they be efficiently computed?

Algorithmic Mechanism and Information Design: In today's digital economy, a significant challenge for online markets and platforms is designing the optimal incentive and information structure for their users. Classical Economic Theory, particularly Mechanism Design, has made strides towards resolving this, but the sheer scale of these markets and the unprecedented volume of data present unique opportunities and complexities. To meet these challenges head-on, the field of Algorithmic Mechanism and Information Design has emerged. It places emphasis on managing the design's computational complexity, leveraging prior knowledge and data robustly,