A0199
Title: Variational inference for spatial correlated failure time data under Bayesian framework
Authors: Yueyao Wang - Zhejiang Gongshang University (China) [presenting]
Abstract: In modern reliability analysis, geographically referenced time-to-event data are often collected. For such reliable data, the spatial dependence on the failure time needs to be properly accounted for in the model. In the literature, spatial random effect models, such as the cumulative exposure model, are often used for analysis with the Bayesian approach as model inference. However, the inference problem is often high dimensional with respect to the number of spatial locations. Consequently, the conventional Markov Chain Monte Carlo (MCMC) methods for sampling the posterior can be very time-consuming when the number of spatial locations is large. Thus, the capability of variational inference (VI) for the inference of spatial survival models is investigated, and a good balance between estimation accuracy and computational efficiency is aimed. Specifically, two divergence metrics, alpha-divergence and the Kullback-Leibler (KL) divergence, are used in the VI methods for the spatial cumulative exposure model. The numerical study compares the MCMC and VI methods under two spatial GPU lifetime data. The comparison results show that the VI method has comparable performance to the MCMC approach but with much more efficient computational time.
