Generating synthetic mobility trajectories using generative modeling


Mobility Datasets are fundamental to evaluate geographic information systems for evaluation and experimental reproducibility. Privacy implications today, however have restricted sharing of such datasets. This has led to the development of synthetic mobility generators, which superficially match mobility characteristics and lack realism. Our approach is to generative realistic synthetic mobility trajectories by appliying machine learning. More specifically, to learn from real mobility datasets and application generative adverserial networks (GANs) to generate synthetic mobility traces.


We will provide you with two mobility datasets collected in Lausanne consisting of over 300 individuals. The two datasets have been collected for two years and 4 months capturing distinct mobility behaviours in the city. Your task will be to apply a GAN based on recurrent neural network (RNN) to generate synthetic a dataset for 50 users. The generated dataset should match the real datasets with respect to several metrics (to be discussed).