In our upcoming Bayesian Working group meeting on Monday (04/14), 1:10 - 2 pm in Snedecor 2113, Andrew Lim will be presenting.

Title: A Survey on Inference and Model Selection for Deep Gaussian Processes

Summary: Deep Gaussian Processes (DGPs) are a deep probabilistic model that extends Gaussian Processes (GPs) by stacking them in layers, similar to the structure of neural networks. Each layer in a DGP models a latent function as a GP, creating a hierarchical model that is able to capture more complex relationships in the data. Though, this complexity comes with caveats as the marginal likelihood, useful for model selection, and exact inference are intractable. In this presentation, we discuss current approaches for inference in DGPs using variational inference and Markov Chain Monte Carlo (MCMC) based methods as well as simulation studies that uses the Laplace Approximation to approximate the marginal likelihood for 2 layer DGPs, for further exploration in model selection.

Look forward to seeing many of you there.