|
Activity Number:
|
270
- Bayesian Data Science and Analytics
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2020 : 1:00 PM to 2:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
|
Abstract #312903
|
|
|
Title:
|
Bayesian Deep Survival Analysis Models for Heterogeneous Electronic Health Record Data
|
|
Author(s):
|
Sounak Chakraborty*
|
|
Companies:
|
University of Missouri, Columbia
|
|
Keywords:
|
deep learning;
deep kernel;
Bayesian;
survival analysis;
big data;
multiplatform data
|
|
Abstract:
|
In this paper we develop hierarchical deep learning models with the help of Deep Exponential Families and embedded deep neural network architecture for survival analysis. A second model we introduce here is a Bayesian accelerated failure time model where the regression function is flexibly modeled using deep kernel of arc-tan nature. Both the models demonstrated superior performance regarding prediction of the survival time and the uncertainty qualification with that. In the deep kernel machine models the finite kernel representation is approximated using random Fourier basis transformation and quasi-MCMC technique. Both of our models are highly efficient in working under big data and multi-platform data integration scenarios.
|
Authors who are presenting talks have a * after their name.
