DeepRV: pre-trained spatial priors for accelerated disease mapping

2025-03-27

AI Summary

Existing VAE-based prior encoding methods (e.g., PriorVAE, πVAE) exhibit strong theoretical foundations but suffer from poor deployability, high training overhead, and limited practical utility. Method: We propose DeepRV, a lightweight decoder-only generative model that introduces the novel “decoder-only prior encoding” paradigm—eliminating the VAE encoder entirely and instead leveraging pretrained spatial statistical priors for efficient disease mapping. Built on NumPyro, DeepRV enables scalable Bayesian inference, achieving MCMC-level accuracy while substantially accelerating variational inference. Contribution/Results: DeepRV provides production-ready APIs, bridging the gap between theoretical modeling and public health practice. Evaluated on real-world tasks—including UK cancer mortality and Zimbabwe HIV prevalence mapping—DeepRV achieves several-fold speedups in inference and reduces parameter estimation error by 32% relative to PriorVAE, closely approaching MCMC benchmark performance.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recently introduced prior-encoding deep generative models (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 accelerates training, and enhances real-world applicability in comparison to current VAE-based prior encoding approaches. Leveraging probabilistic programming frameworks (e.g., NumPyro) for inference, DeepRV achieves significant speedups while also improving the quality of parameter inference, closely matching full MCMC sampling. We showcase its effectiveness in process emulation and spatial analysis of the UK using simulated data, gender-wise cancer mortality rates for individuals under 50, and HIV prevalence in Zimbabwe. To bridge the gap between theory and practice, we provide a user-friendly API, enabling scalable and efficient Bayesian inference.

Problem

• Accelerates disease mapping using lightweight deep generative models
• Improves parameter inference quality compared to VAE-based approaches
• Bridges theory-practice gap with user-friendly API for Bayesian inference

Innovation

• Lightweight decoder-only approach for faster training
• Uses probabilistic programming for efficient inference
• User-friendly API for practical Bayesian inference