Recurrent Neural Network Models for Route Planning in Road Networks
Summary: The paper describes an evaluative analysis of applying recurrent neural networks for solving the shortest path problem in real road networks. An RNN-based graph search algorithm is proposed. The algorithm is trained on a real road network to generate shortest paths by duration. The algorithm is tested on a real road network. The results show that RNNs can yield optimal routes. Yet, the algorithm is not robust enough - integrating the proposal with existing classical optimal path search methods is suggested.
Supervisor: Linas Petkevičius