arXiv:2607.12666 2026 Sampling 1 ideas extracted · analyzed Aug 30, 2026
The paper provides a path-level large-deviation description for empirical scheduled-arrival trajectories, where deterministic appointment times are perturbed by independent unpunctuality variables. The transferable asset is an explicit variational rate function on entire paths, together with exponential tilting by linear path functionals; this can generate rare congestion trajectories without relying on extremely imbalanced Monte Carlo data. A practical neural-network use is rare-event augmentation for queue predictors, event-sequence models, or world models, with likelihood-ratio weighting so that tilted examples do not corrupt the nominal training distribution.
Train a neural queue or event-sequence predictor using trajectories generated under an exponentially tilted scheduled-arrival law that makes rare overloads common. Reweight each tilted trajectory by its likelihood ratio, while optionally oversampling the rare-event subset to improve prediction of tail behavior.