Abstract
We study machine learning-based assistants that support coordination betweenhumans in congested facilities via congestion forecasts. In our theoreticalanalysis, we use game theory to study how an assistant's forecast thatinfluences the outcome relates to Nash equilibria, and how they can be reachedquickly in congestion game-like settings. Using information theory, weinvestigate approximations to given social choice functions under privacyconstraints w.r.t. assistants. And we study dynamics and training for aspecific exponential smoothing-based assistant via a linear dynamical systemsand causal analysis. We report experiments conducted on a real congestedcafeteria with about 400 daily customers where we evaluate this assistant andprediction baselines to gain further insight.