Skip to content

Sequential Inference for Gaussian Processes: A Signal Processing Perspective

Home / Papers / Sequential Inference for Gaussian Processes: A Signal Processing Perspective

Sequential Inference for Gaussian Processes: A Signal Processing Perspective

Authors: Daniel Waxman, Fernando Llorente, Petar M. Djurić Date: 2026-04-30 Paper ID: openalex:2604.28163

Summary

This paper provides a systematic, tutorial-style overview of Gaussian Processes (GPs) specifically for sequential, incremental, and streaming inference tasks. By reframing GP methodologies through a signal processing lens, the authors bridge the gap between classical statistical signal processing and modern machine learning paradigms. The work serves as a foundational guide for researchers and practitioners, offering a coherent roadmap for implementing sequential GP models in applications like state-space modeling, adaptive sensing, and time series forecasting.

Key Contributions

  • Provides a comprehensive tutorial-style overview of Gaussian Processes (GPs) focused on sequential, incremental, and streaming inference.
  • Bridges methodological gaps between classical signal processing and modern machine learning frameworks for temporal data.
  • Organizes recent advancements in sequential GP inference into a unified roadmap for practical deployment in real-world systems.

Open Questions & Future Work

Archivist Review

The paper is a high-level tutorial review of Gaussian Processes for sequential inference. As it is a survey rather than a novel method proposal, no new concepts are introduced. However, the identified open questions regarding non-stationarity in high dimensions and decentralized scalability are substantial, foundational bottlenecks for the field of sequential probabilistic modeling, and thus warrant inclusion in the vault.

Approved Open Questions

  • Scalable nonstationary sequential GPs: This is a fundamental bottleneck for deploying GP-based sequential estimation in real-world applications where the underlying signal dynamics are inherently nonstationary and high-dimensional. Bridging this gap is essential to make GPs competitive with modern neural architectures.
  • Scalability in decentralized GPs: As distributed sensing and multi-agent systems become more ubiquitous, the ability to perform robust, collaborative probabilistic inference in large-scale networks will be critical for intelligent and adaptive system control.

Metadata & Links

url
https://arxiv.org/abs/2604.28163
paper_id
2604.28163
paper_source
openalex
domain
time-series
tags
time-seriesbayesian-inference
architectures
datasets
concept_slugs
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-05-03T06:03:07Z