Abstract
dc:description.abstractThis dissertation studies problems that standard theories cannot solve well and theories with learning algorithms can better solve. Across three essays, I study how information designers and mechanisms achieve approximate optimality while being robust to real-world challenges, such as unknown users’ beliefs, unknown state distribution, and the agent's inability to best respond. The first chapter studies the information design problem when an information designer, such as an online platform, does not know the beliefs of its receivers. The designer interacts with a sequence of myopic receivers, whose prior beliefs over states coincide with the true distribution of states. We design learning algorithms so that the information designer can learn that prior belief through repeated interactions. Our learning algorithms achieve no regret relative to optimality for the known prior at a fast speed, achieving a tight regret bound $\Theta(\log T)$ in general and a tight regret bound $\Theta(\log \log T)$ in an important binary-action case. The second chapter considers the information design problem in which the designer does not know the prior belief of a strategic (long-lived) receiver, who may take deceptive actions to manipulate what the designer learns. We construct learning algorithms enabling the designer to learn the receiver’s belief through repeated interactions. Our learning algorithms are robust to the receiver’s strategic manipulation of the learning process of the designer. We study regret relative to two benchmarks to measure the performance of the algorithms. The static benchmark is $T$ times the single-period optimum for the designer under the receiver's belief. The dynamic benchmark, which is stronger, characterizes global dynamic optimality for the designer under the receiver's belief. Our learning algorithms allow the designer to achieve no regret against both benchmarks at fast speeds of O(\log2 T /T). The third chapter studies a mechanism design problem in which an investing bidder (investor) cannot best respond but is able to use a no-regret learning algorithm to adapt his investment across repeated interactions with a dynamic environment. We analyze how the welfare performance guarantee of approximation allocation algorithms extends from a static environment to a dynamic environment. The performance of an approximation allocation algorithm is measured by the approximation ratio between the welfare induced by it and a benchmark. We consider two benchmarks: the best-in-hindsight benchmark, which is the investor's globally optimal utility from a best-in-hindsight investment across time, and a stronger time-varying benchmark, which consists of the investor's optimal single-period utility from the optimal investment in each period. First, against the best-in-hindsight benchmark, the approximation ratio in the dynamic environment coincides with that in the static benchmark. Second, against the time-varying benchmark, we characterize tight upper and lower bounds on the approximation ratio in the dynamic environment compared to that in the static environment. Bridging mechanism design and online learning, we show how welfare guarantees of approximation allocation algorithms can be maintained robustly even when a bidder cannot best respond but learn his investment strategy in a complex and uncertain environment.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Economics
- Grantor
- Boston University
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Li, Ce
- Advisor dc:contributor.advisor
-
- Lipman, Barton
- ORCID dc:identifier.orcid
Subjects
dc:subject × 7Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/2144/53432