Skip to content

Generative Modeling of Approximately Periodic Time Series by a Posterior-Weighted Gaussian Process

Home / Papers / Generative Modeling of Approximately Periodic Time Series by a Posterior-Weighted Gaussian Process

Generative Modeling of Approximately Periodic Time Series by a Posterior-Weighted Gaussian Process

Authors: Elias Reich, Saverio Messineo, S. Huber Date: 2026-05-13 Paper ID: openalex:2605.13150

Summary

This paper addresses the challenge of modeling approximately periodic time series, where signals exhibit repetitive trajectories with varying duration, amplitude, and dynamics. The authors propose a stochastic generative model based on a Gaussian Process with a posterior-modulated kernel, effectively decoupling intra-repetition structure from inter-repetition variability. This approach allows for the generation of synthetic trajectories that maintain consistent mean structures while introducing smooth, realistic inter-repetition fluctuations.

Key Contributions

  • Introduces a posterior-weighted Gaussian Process that enables generative modeling of approximately periodic time series.
  • Proposes a two-stage kernel construction to decouple intra-repetition structural regularity from inter-repetition variability.
  • Demonstrates the ability to synthesize realistic trajectories that preserve global structure while accounting for variations in duration, amplitude, and dynamics.

Open Questions & Future Work

Key Concepts

  • posterior-weighted-gaussian-process: A Gaussian Process model that uses a modulated kernel to generate approximately periodic time series by decoupling local trajectory structure from inter-repetition variation.

Archivist Review

I approved the core ‘Posterior-Weighted Gaussian Process’ mechanism as it provides a distinct, reusable approach to decomposing time series periodicity. I also approved the open question regarding its scalability, as this represents a significant bottleneck for applying covariance-modulated GPs to long-horizon, high-repetition industrial data. The other candidate question was rejected as being too broad and dependent on general model improvement.

Approved Concepts

  • Posterior-Weighted Gaussian Process: The paper introduces this as the core mechanism to decouple intra-repetition structure from inter-repetition variability in approximately periodic data.

Approved Open Questions

  • Scalable Inference for Modulated GP Covariance: The current method’s computational complexity limits its application to datasets with a large number of repetitions; addressing this is critical for the model’s scalability in industrial contexts.

Metadata & Links

url
https://arxiv.org/abs/2605.13150
paper_id
2605.13150
paper_source
openalex
domain
time-series
tags
architectures
datasets
concept_slugs
posterior-weighted-gaussian-process
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-05-16T06:01:30Z