Enhancing the Adaptiveness of Gaussian Process Regression based on Power Spectral Density

摘要

In many years, Gaussian process was popularly utilized in many research areas such as signal processing, data communications and image processing, etc. Unlike other techniques which try to determine all of the parameters of system model, Gaussian process adapts these parameters to reflect the actual underlying model. Because of that, this approach can be explicitly addressed as a non-parametric methodology. As a comparison to other well-known methods, Gaussian process regression (GPR) possesses much better performance in terms of precision and versatility. However, this technique does have some drawbacks. One of them is the adaptiveness to the complex data. In this research, we would like to introduce a novel solution based on power spectral density to adapt the model for better accuracy.