.png){kind=icon}
Return
An efficient parameter inference method for dynamical networks
.png){kind=icon}
PRE
AI
DOI：10.1016/j.physd.2026.135391.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Research in complex networks often focuses on parameter inference on the basis of dynamical data. Difficulties usually arise from complex network structures and nonlinear interactions between individual nodes and their neighbours. We present an efficient method to address the challenge. According to dynamical data and specific inference requests, a loss function is set up. Benefiting from the definite network topology and nodal equations of a known dynamical model, the unknown physical parameters are ultimately optimized and identified by continuously minimizing a loss function with the help of the diverse and efficient techniques of artificial neural networks, including gradient-based learning, automatic differentiation and optimization algorithms. Numerical simulations demonstrate that this framework can accurately infer critical couplings and other system parameters. Besides offering high-precision and quick inference results, our method is robust to network size, structure, and parameter variations. Meanwhile, the method is capable of dealing with both discrete and continuous systems. In addition, a scaling law of critical values is derived from dynamical analysis, and the reliability of our method is confirmed.
Keywords:
Parameter inference
Artificial neural network
Complex network
Dynamical data
Loss function
Journal
P
IF:
2.9
Papers:
142
Citations:
0
Organization
Cited Papers
No cited papers available