Scalable Probabilistic Routes

Inference and prediction of routes have become interest over the past decade owing to a dramatic increase in package delivery ride-sharing services. Given underlying combinatorial structure incorporation probabilities, route involves techniques from both formal methods machine learning. One promising approach for predicting uses decision diagrams that are augmented with probability values. However, effectiveness this depends on size compiled diagrams. The scalability is limited its empirical runtime space complexity. In work, our contributions two-fold: first, we introduce relaxed encoding linear number variables respect vertices road network graph significantly reduce resultant Secondly, instead stepwise sampling procedure, propose single pass sampling-based prediction. evaluations arising real-world network, demonstrate resulting system achieves around twice quality suggested while being an order magnitude faster compared state-of-the-art.