2024/09/18 by Meng, Qiuxian, Yongyou Zhang, Zhang, Yongyou
Computer Science · #FOS: Computer and information sciences #Gaussian Processes and Bayesian Inference #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning in Healthcare #Time Series Analysis and Forecasting
paper · pdf · doi:10.48550/arxiv.2409.12301
openalex publication_date 2024/09/18 · openalex created_date 2025/10/24 · openalex updated_date 2026/07/28
Gaussian processes (GPs) are Bayesian nonparametric models for function\napproximation with principled predictive uncertainty estimates. Deep Gaussian\nprocesses (DGPs) are multilayer generalizations of GPs that can represent\ncomplex marginal densities as well as complex mappings. As exact inference is\neither computationally prohibitive or analytically intractable in GPs and\nextensions thereof, some existing methods resort to variational inference (VI)\ntechniques for tractable approximations. However, the expressivity of\nconventional approximate GP models critically relies on independent inducing\nvariables that might not be informative enough for some problems. In this work\nwe introduce amortized variational inference for DGPs, which learns an\ninference function that maps each observation to variational parameters. The\nresulting method enjoys a more expressive prior conditioned on fewer input\ndependent inducing variables and a flexible amortized marginal posterior that\nis able to model more complicated functions. We show with theoretical reasoning\nand experimental results that our method performs similarly or better than\nprevious approaches at less computational cost.\n