Inverse Learning of Network Equilibrium

Reading Rules from Routes

Conventional traffic modeling starts with assumptions about demand and road costs, then computes how travelers distribute themselves across a network. In practice, planners often face the reverse situation: the network and some traffic observations are known, but key behavioral or congestion parameters are not.

The opening acts put the forward and inverse questions side by side. Several observed traffic states then help narrow down the unknown rules that could have produced them. Once those rules are learned, the video adds a new road and asks what traffic might look like under that changed network. This counterfactual is the reason inverse learning matters. Observations are not merely used to reproduce yesterday's flows; they are used to construct a model that can reason about an intervention that has not yet happened. The quality of that prediction still depends on whether the observed states contain enough information to identify the relevant behavior.