摘要:
The generalized polynomial chaos (gPC) are widely used as surrogate models in Bayesian inference to speed up the Markov chain Monte Carlo simulations. However, the use of gPC-surrogates introduces model errors that may severely distort the estimate of the posterior distribution. In this talk, we present an adaptive procedure to construct an adaptive gPC-surrogate. The key idea is to refine the surrogate over a sequence of samples adaptively so that the surrogate is much more accurate in the posterior region. We then introduce an adaptive surrogate modeling approach based on deep neural networks to handle problems with high dimensional parameters.
报告人:
周涛,中国科学院数学与系统科学研究院副研究员。曾于瑞士洛联邦理工大学从事博士后研究。主要研究方向为不确定性量化、时间并行算法以及随机最优控制等。在SIAM Review、SINUM、Math. Comput.等期刊发表论文50余篇。2016年获中国工业与应用数学学会青年科技奖,2018年获国家自然科学基金委“优秀青年科学基金”资助。2017年起担任国际不确定性量化期刊International Journal for UQ副总编,并同时担任国际科学计算权威期刊SIAM J. Sci. Comput.及Commun. Comput. Phys.等多个国际期刊编委。2018起担任国防科工局科学挑战专题领域一“复杂系统模型不确定性评定方法”首席科学家
时间:2020年8月25日周一上午9:00-10:00
腾讯会议:310 917 012
欢迎广大师生参加!