B0332
Title: BKTR - Bayesian kernelized tensor regression: Application to bike-sharing demand modeling
Authors: Aurelie Labbe - HEC Montreal (Canada) [presenting]
Mengying Lei - McGill (Canada)
Lijun Sun - McGill (Canada)
Abstract: As a regression technique in spatial statistics, the spatiotemporally varying coefficient model (STVC) is an important tool for discovering nonstationary and interpretable response-covariate associations over both space and time. However, it is difficult to apply STVC for large-scale spatiotemporal analyses due to its high computational cost. To address this challenge, the spatiotemporally varying coefficients are summarized using a third-order tensor structure and propose to reformulate the spatiotemporally varying coefficient model as a special low-rank tensor regression problem. The low-rank decomposition can effectively model the global patterns of large data sets with a substantially reduced number of parameters. To further incorporate the local spatiotemporal dependencies, Gaussian process (GP) priors are used on the spatial and temporal factor matrices. The overall framework is referred to as Bayesian kernelized tensor regression (BKTR). For model inference, an efficient Markov chain Monte Carlo (MCMC) algorithm is developed, which uses Gibbs sampling to update factor matrices and slice sampling to update kernel hyperparameters. Extensive experiments on both synthetic and real-world data sets are conducted on bike-sharing demand, and the results confirm the superior performance and efficiency of BKTR for model estimation and parameter inference.
