|
Activity Number:
|
302
- Advances in Bayesian Computation
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
|
Abstract #306753
|
Presentation
|
|
Title:
|
Statistical and Computational Guarantees for Variational Boosting
|
|
Author(s):
|
Biraj Guha* and Debdeep Pati and Anirban Bhattacharya
|
|
Companies:
|
Texas A & M University and Texas A&M University and TAMU
|
|
Keywords:
|
Variational Inference;
Variational Boosting;
Bayesian computing
|
|
Abstract:
|
In this work, we provide theoretical guarantees for the convergence of variational approximation to the true posterior for a wider class of variational family than considered in the literature. In particular we show that the Kullback-Leibler divergence between the variational estimate and the true posterior is of the optimal order under appropriate smoothness and identifiability conditions and exponential tail behavior of the likelihood. In addition, we provide computational guarantees for a version of the variational boosting algorithm that ensures that the iterates converge to the true posterior with a high probability under the true data generation mechanism.
|
Authors who are presenting talks have a * after their name.
