Please use this identifier to cite or link to this item:
https://scholarbank.nus.edu.sg/handle/10635/165047
Title: | METAMODELING AND OPTIMIZATION WITH GAUSSIAN PROCESS MODELS FOR STOCHASTIC SIMULATIONS | Authors: | WANG SONGHAO | Keywords: | Gaussian process model, multi-response, stochastic metamodeling, simulation optimization, Bayesian Optimization, convergence | Issue Date: | 21-Aug-2019 | Citation: | WANG SONGHAO (2019-08-21). METAMODELING AND OPTIMIZATION WITH GAUSSIAN PROCESS MODELS FOR STOCHASTIC SIMULATIONS. ScholarBank@NUS Repository. | Abstract: | The Gaussian process (GP) model and GP-based Bayesian optimization (BO) algorithms have been widely used in computer simulations. This thesis proposes three pieces of work to address some challenges and enhance their applications for stochastic simulations. Firstly, we propose a multi-response stochastic GP model to improve the prediction of primary responses, corrupted with noise, by learning from some correlated and more accurate auxiliary response. Secondly, we extend the application of BO to optimizing the output distribution quantiles by leveraging the accurate information from the informative lower quantiles, which are easier to optimize compared with the quantiles at the objective level, to find the optimums faster. Finally, we propose a general class of BO, which works with a broad range of sampling criteria with convergence guarantee. A specific algorithm is then developed with finite-time regret bound, which, to the best of our knowledge, is the first for stochastic BO algorithms. | URI: | https://scholarbank.nus.edu.sg/handle/10635/165047 |
Appears in Collections: | Ph.D Theses (Open) |
Show full item record
Files in This Item:
File | Description | Size | Format | Access Settings | Version | |
---|---|---|---|---|---|---|
WangS.pdf | 2.39 MB | Adobe PDF | OPEN | None | View/Download |
Google ScholarTM
Check
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.