题目：Iterative ensemble smoothers for large-scale subsurface inverse problems
报告人：Dean Oliver （NORCE Norwegian Research Centre AS）
邀请人：雷荔傈 教授
时间：2019年4月1日，上午10：30
地点：bwin必赢D103报告厅
摘要：The key features of inverse problems applied to subsurface flow models are that the number of parameters are large (∼ 105 - 107) and that the relationships between observations and model parameters are highly nonlinear and nonlocal. The cost of evaluating the likelihood is relatively high (15 min to 1 day). To assimilate data into these types of models, we typically generate samples of model parameters and observation error from the prior distributions, then minimize a stochastic cost function of model parameters. When adjoint methods are used to compute the gradient of the cost function with respect to model parameters, the distribution of the resulting realizations is often very close to the actual posterior distribution, even when the posterior has multiple modes. For very large subsurface flow problems, however, we use iterative ensemble smoothers to assimilate the data which results in additional approximations to the posterior. I will illustrate methodology with applications to both small toy problems and a real subsurface flow problem.
欢迎老师同学踊跃参加！