Online method improves warping parameter updates in gaussian processes
Online Gradient Computation for Warping Gaussian Process Transformations
Machine Learning
Summary
Some models called warped Gaussian processes help computers understand weird data that doesn’t fit usual assumptions. The paper shows how to quickly and accurately update key parts of these models as new data comes in, without waiting for breaks or losing precise math details. The authors created a step-by-step way to calculate changes needed to keep the model working well in real time. This improvement makes the model better at adapting to new information as it arrives.
What this means in practice
- •For signal processing engineers: Update models that analyze irregular data in real time without losing accuracy on parameter changes.
- •For financial data analysts: Keep predictive models current by efficiently refining transformation parameters as new, non-standard financial data streams in.
Authors
Emilio Ruiz-Moreno, Konstantinos Slavakis, Baltasar Beferull-Lozano
Abstract
Warped Gaussian processes (GPs) handle non-Gaussian observations by mapping them into a latent standard GP via a parametric transformation called warping. Existing streaming variants, however, either optimize the warping parameters periodically or sacrifice analytical tractability for a higher model capacity. To bridge this gap, we show that the gradient of the instantaneous negative log-likelihood of a warped GP admits an exact recursive computation. Based on this result, we propose a novel online method for warped GPs that jointly updates the latent GP moments and optimizes the warping parameters.