STACI: Spatio-Temporal Aleatoric Conformal Inference

2025-05-27

AI Summary

This paper addresses the challenges of inaccurate aleatoric uncertainty quantification and poor scalability in spatiotemporal field estimation. Methodologically, it introduces a unified framework integrating variational Bayesian neural networks with a novel spatiotemporal conformal inference scheme. Specifically: (1) it proposes the first scalable, nonstationary spatiotemporal Gaussian process neural approximation, circumventing restrictive kernel assumptions and cubic computational complexity inherent in conventional Gaussian processes; (2) it designs a spatiotemporal dependency-aware conformal prediction algorithm that guarantees finite-sample statistical validity of prediction intervals. The framework enables GPU-accelerated training on datasets with over one million observations. Empirically, it achieves prediction intervals with coverage strictly matching user-specified confidence levels—outperforming state-of-the-art Gaussian process and spatiotemporal deep learning baselines in both calibration and efficiency.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks

Abstract

Fitting Gaussian Processes (GPs) provides interpretable aleatoric uncertainty quantification for estimation of spatio-temporal fields. Spatio-temporal deep learning models, while scalable, typically assume a simplistic independent covariance matrix for the response, failing to capture the underlying correlation structure. However, spatio-temporal GPs suffer from issues of scalability and various forms of approximation bias resulting from restrictive assumptions of the covariance kernel function. We propose STACI, a novel framework consisting of a variational Bayesian neural network approximation of non-stationary spatio-temporal GP along with a novel spatio-temporal conformal inference algorithm. STACI is highly scalable, taking advantage of GPU training capabilities for neural network models, and provides statistically valid prediction intervals for uncertainty quantification. STACI outperforms competing GPs and deep methods in accurately approximating spatio-temporal processes and we show it easily scales to datasets with millions of observations.

Problem

• Addresses scalability issues in spatio-temporal Gaussian Processes
• Improves simplistic covariance assumptions in deep learning models
• Provides valid prediction intervals for uncertainty quantification

Innovation

• Variational Bayesian neural network for non-stationary GP
• Spatio-temporal conformal inference algorithm
• GPU-accelerated scalable training for large datasets