About the event
Using the London Underground as our primary example, we will discuss two main lines of work. In the first, we discuss a practical end-to-end model to infer changes in passenger behaviour based on unplanned disruption. This model takes into account user-level data to account for heterogeneous choice inferred from smart card outputs. In the second line of work, we describe an abstract view of causal models for complex systems when some records of shocks are available along plentiful observational data. We reduce this problem to distribution learning in a structured domain, and discuss its practical implications. Joint work with Nicolo Colombo, Edoardo Airoldi and Soong Kang
