Yahoo! JAPAN R&D

The following paper won the Excellent Paper Award at IPSJ-SIGUBI 60th workshop (external link).

Urban Dynamics Prediction Sharing GPS Logs between Cities by HDP Regression Mixture Masamichi Shimosaka (Tokyo Tech), Yuta Hayakawa (Tokyo Tech), Kota Tsubouchi

We predict urban dynamics in a city using only the GPS logs in the city, and the existing models cause instability in the learning and the model cannot consider the relationship between cities. To solve this issue, we proposed the hierarchical Dirichlet process (HDP) regression mixture using HDP technique for the prior of the predictive model. The proposed model can predict urban dynamics much stable for each cities sharing the dataset and the parameters between cities.