HAYSTAC (Hidden Activity Signal and Trajectory Anomaly Characterization)

Research Members: Albert Cao, Giuseppe Perona

The goal of the Hidden Activity Signal and Trajectory Anomaly Characterization (HAYSTAC) is to develop a generative model that produces complete trajectories of stay locations given sparse Location-Based Service (LBS) data. We leverage the state-of-the-art transformers, a deep learning architecture capable of capturing both the short term and long term dependencies in sequences of inputs. This allows us to impute the missing stay locations considering the relative correlation among different locations. We construct domain knowledge-based loss functions and inject individual heterogeneity by varying the ground truth label in the loss. Our model is capable of producing complete sequences of stay location while preserving the universal human mobility laws.

Publications

HuMNet Lab