To date, Bayesian experimental design (BED) has largely studied how a single adaptive policy should choose experiments. Yet many real-world settings involve multiple designers making experimental decisions independently, including laboratories, field teams, and decentralised robots. Even when individual designers choose informative experiments, they may collectively acquire redundant information. In this talk, we introduce multi-designer Bayesian experimental design (MD-BED) and explore the coordination problem that arises when designers share an inferential objective but act using only their local histories. We show how this problem can be characterised information-theoretically, and how centralised training with decentralised execution (CTDE) can improve information acquisition and encourage designer specialisation under heterogeneous experimental costs, without requiring communication during execution.

Marcel Hedman is a DPhil candidate in Statistical Machine Learning at the University of Oxford, supported by the StatML fellowship and supervised by Professor Chris Holmes and Professor Tom Rainforth. His research focuses on sequential decision making, with particular interests in multi-agent reinforcement learning, world modelling and Bayesian experimental design.