TY - GEN
T1 - Variational inference for Gaussian process models for survival analysis
AU - Kim, Minyoung
AU - Pavlovic, Vladimir
N1 - Funding Information: MK is supported by National Research Foundation of Korea (NRF-2016R1A1A1A05921948).
PY - 2018
Y1 - 2018
N2 - Gaussian process survival analysis model (GP-SAM) was recently proposed to address key deficiencies of the Cox proportional hazard model, namely the need to account for uncertainty in the hazard function modeling while, at the same time, relaxing the time-covariates factorized assumption of the Cox model. However, the existing MCMC inference algorithms for GPSAM have proven to be slow in practice. In this paper we propose novel and scalable variational inference algorithms for GP-SAM that reduce the time complexity of the sampling approaches and improve scalability to large datasets. We accomplish this by employing two effective strategies in scalable GP: i) using pseudo inputs and ii) approximation via random feature expansions. In both setups, we derive the full and partial likelihood formulations, typically considered in survival analysis settings. The proposed approaches are evaluated on two clinical and a divorce-marriage benchmark datasets, where we demonstrate improvements in prediction accuracy over the existing survival analysis methods, while reducing the complexity of inference compared to the recent state-of-the-art MCMC-based algorithms.
AB - Gaussian process survival analysis model (GP-SAM) was recently proposed to address key deficiencies of the Cox proportional hazard model, namely the need to account for uncertainty in the hazard function modeling while, at the same time, relaxing the time-covariates factorized assumption of the Cox model. However, the existing MCMC inference algorithms for GPSAM have proven to be slow in practice. In this paper we propose novel and scalable variational inference algorithms for GP-SAM that reduce the time complexity of the sampling approaches and improve scalability to large datasets. We accomplish this by employing two effective strategies in scalable GP: i) using pseudo inputs and ii) approximation via random feature expansions. In both setups, we derive the full and partial likelihood formulations, typically considered in survival analysis settings. The proposed approaches are evaluated on two clinical and a divorce-marriage benchmark datasets, where we demonstrate improvements in prediction accuracy over the existing survival analysis methods, while reducing the complexity of inference compared to the recent state-of-the-art MCMC-based algorithms.
UR - http://www.scopus.com/inward/record.url?scp=85059367688&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85059367688&partnerID=8YFLogxK
M3 - Conference contribution
T3 - 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018
SP - 435
EP - 445
BT - 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018
A2 - Globerson, Amir
A2 - Silva, Ricardo
PB - Association For Uncertainty in Artificial Intelligence (AUAI)
T2 - 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018
Y2 - 6 August 2018 through 10 August 2018
ER -