Gaussian process regression with additive periodic kernels for two-body interaction analysis in coupled phase oscillators
Gaussian process regression with additive periodic kernels for two-body interaction analysis in coupled phase oscillators
We propose a Gaussian process regression framework with additive periodic kernels for the analysis of two-body interactions in coupled oscillator systems. While finite-order Fourier expansions determined by Bayesian methods can still yield artifacts such as a high-amplitude, high-frequency vibration, our additive periodic kernel approach has been demonstrated to effectively circumvent these issues. Furthermore, by exploiting the additive and periodic nature of the coupling functions, we significantly reduce the effective dimensionality of the inference problem. We first validate our method on simple coupled phase oscillators and demonstrate its robustness to more complex systems, including Van der Pol and FitzHugh-Nagumo oscillators, under conditions of biased or limited data. We next apply our approach to spiking neural networks modeled by Hodgkin-Huxley equations, in which we successfully recover the underlying interaction functions. These results highlight the flexibility and stability of Gaussian process regression in capturing nonlinear, periodic interactions in oscillator networks. Our framework provides a practical alternative to conventional methods, enabling data-driven studies of synchronized rhythmic systems across physics, biology, and engineering.
Ryosuke Yoneda、Haruma Furukawa、Daigo Fujiwara、Toshio Aoyagi
物理学
Ryosuke Yoneda,Haruma Furukawa,Daigo Fujiwara,Toshio Aoyagi.Gaussian process regression with additive periodic kernels for two-body interaction analysis in coupled phase oscillators[EB/OL].(2025-04-12)[2025-04-27].https://arxiv.org/abs/2504.09110.点此复制
评论