DeepRV: pre-trained spatial priors for accelerated disease mapping.

Jhonathan Navott, Daniel Jenson, Seth Flaxman, Elizaveta Semenova

Keywords: Bayesian Inference, Gaussian Processes, spatial modeling, variational autoencoders

TL;DR: Decoder-only deep generative model for pre-trained spatial priors, enabling accelerated modeling and inference in disease mapping.

Abstract:

Recently introduced deep generative priors (e.g., PriorVAE, $\pi$VAE, and PriorCVAE) have emerged as powerful tools for scalable Bayesian inference by emulating complex stochastic processes like Gaussian processes (GPs).
However, these methods remain largely a proof-of-concept and inaccessible to practitioners. We propose DeepRV, a lightweight, decoder-only approach that reduces the number of parameters by 66\%, accelerates training, and enhances real-world applicability in comparison to current VAE-based approaches. Leveraging probabilistic programming frameworks (e.g., NumPyro), DeepRV achieves an order-of-magnitude speedup while maintaining robust performance. We showcase its effectiveness in GP emulation and spatial analysis of the UK using simulated data and cancer mortality rates (Work in progress). To bridge the gap between theory and practice, we provide a user-friendly API, enabling scalable and efficient Bayesian inference.

Submission Number: 66