Neural simulation-based inference is a recent paradigm shift in statistical science that uses simulations from a statistical model to train a neural network for statistical inference. These neural models serve as fast and accurate amortized surrogates for classical inferential quantities, such as point estimators, confidence sets, likelihoods, posteriors or conditional distributions, whose evaluation tends to be computationally intensive or wholly intractable for many complex statistical models. As such, this approach holds the potential for massive scale-up and increased fidelity of statistical inference in complex real-world settings which have previously relied on crude approximations or overly simplified modeling. In this talk, I will first give an overview of some of the key ideas in neural simulation-based inference. I will then describe our work on neural likelihood inference motivated by complex inferential problems in spatial and spatio-temporal statistics. Specific topics covered include scalable Gaussian process modeling of Earth science observations and circumventing intractable likelihoods in spatial extremes models and spatial transport error models, with an application to constraining aerosol climate parameters.

Department reception in 450 West Hall after the seminar at 11:00 AM