Gary Qiurui

Gary Qiurui Ma

Modern online platforms such as marketplaces, ride-hailing services, and food-delivery systems serve a dual role: they are both markets where participants interact and transact, and operators that design and govern how these markets function. These platforms connect multiple sides, for example buyers, sellers, and couriers, facilitating access that would otherwise be difficult to achieve. By setting the rules of the market, platforms determine who participates, how interactions take place, and how value is created and distributed. In response to these rules, participants may behave strategically, deciding whether to join the platform and which transactions to pursue. This thesis studies how platform design affects market outcomes through three key levers: pricing that determines participants' gains when operating on a platform; matching that governs which interactions are feasible among participants; and bundling that shapes the structure of supply when the platform itself acts as a market participant. Across these levers, the goal in this thesis is to understand how platforms can be designed to balance platform profitability with overall market welfare. The first part of this thesis studies pricing, including both the commission fees that participants pay to a platform and the prices associated with each transaction. The second part of this thesis studies matching. By shaping recommendation systems and consumer search, platforms influence which transactions take place. The third part of this thesis analyzes bundling. As a marketplace operator, a platform may be able to source products from sellers and offer them as bundled packages to buyers. Collectively, this thesis shows how pricing, matching, and bundling serve as complementary design levers through which platforms can shape market outcomes.

Junning Huang, Sirui Xie, Jiankai Sun et al.

Autonomous driving systems have a pipeline of perception, decision, planning, and control. The decision module processes information from the perception module and directs the execution of downstream planning and control modules. On the other hand, the recent success of deep learning suggests that this pipeline could be replaced by end-to-end neural control policies, however, safety cannot be well guaranteed for the data-driven neural networks. In this work, we propose a hybrid framework to learn neural decisions in the classical modular pipeline through end-to-end imitation learning. This hybrid framework can preserve the merits of the classical pipeline such as the strict enforcement of physical and logical constraints while learning complex driving decisions from data. To circumvent the ambiguous annotation of human driving decisions, our method learns high-level driving decisions by imitating low-level control behaviors. We show in the simulation experiments that our modular driving agent can generalize its driving decision and control to various complex scenarios where the rule-based programs fail. It can also generate smoother and safer driving trajectories than end-to-end neural policies.