Supervisor

Suitable for

Abstract

In information design, a more-informed player (sender) influences a less-informed decision-maker (receiver) by signalling information about the state of the world. The problem for the sender is to compute an optimal signalling strategy, which leads to the receiver taking actions that benefit the sender. Dynamic information design, as a new frontier of information design, generalises the one-shot framework to dynamic settings that are modelled based on Markov decision processes. The goal of the project is to study several variants of the dynamic information design problem and it can be approached from both theoretical and empirical perspectives. Theoretically, the focus is on determining the computational complexity of the optimal information design in different dynamic settings and developing algorithms. A background in computational complexity and algorithm design is beneficial. Practically, the objective is to apply existing algorithms to novel applications, such as traffic management or board games, and to develop algorithms that work effectively in real-world scenarios using state-of-the-art methods. Knowledge in Markov Decision Processes, stochastic/sequential games, and Reinforcement Learning is preferred.

Related work:

J. Gan, R. Majumdar, G. Radanovic, A. Singla. Bayesian persuasion in sequential decision-making. AAAI '22

E. Kamenica, M. Gentzkow. Bayesian persuasion. American Economic Review, 2011

S. Dughmi. Algorithmic information structure design: a survey. ACM SIGecom Exchanges 15.2 (2017): 2-24.

Wu, J., Zhang, Z., Feng, Z., Wang, Z., Yang, Z., Jordan, M. I., & Xu, H. (2022). Sequential Information Design: Markov Persuasion Process and Its Efficient Reinforcement Learning. arXiv preprint arXiv:2202.10678.