Abstract:Discovering governing equations of complex network dynamics is a fundamental challenge in contemporary science with rich data, which can uncover the mysterious patterns and mechanisms of the formation and evolution of complex phenomena in various fields and assist in decision-making. In this work, we develop a universal computational tool that can automatically, efficiently, and accurately learn the symbolic changing patterns of complex system states by combining the excellent fitting ability from deep learning and the equation inference ability from pre-trained symbolic regression. We conduct intensive experimental verifications on more than ten representative scenarios from physics, biochemistry, ecology, epidemiology, etc. Results demonstrate the outstanding effectiveness and efficiency of our tool by comparing with the state-of-the-art symbolic regression techniques for network dynamics. The application to real-world systems including global epidemic transmission and pedestrian movements has verified its practical applicability. We believe that our tool can serve as a universal solution to dispel the fog of hidden mechanisms of changes in complex phenomena, advance toward interpretability, and inspire more scientific discoveries.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: how to automatically, efficiently and accurately discover the control equations from complex network dynamics. Specifically, the authors have developed a general - purpose computational tool to automatically learn the change patterns of complex system states by combining the excellent fitting ability of deep learning and the equation - reasoning ability of pre - trained symbolic regression.
### Problem Background
1. **Dynamic Changes in Complex Systems**
- Dynamic changes in complex systems are usually described by a set of nonlinear differential equations, for example:
\[
\dot{X}(t)=f(X(t), A, t)
\]
where \( X(t)\in\mathbb{R}^{N\times d} \) represents the state of the system at time \( t \), \( N \) and \( d \) represent the number of system components (nodes) and the state dimension respectively, and \( A \) represents additional information beyond the system state, such as topological interactions between components.
2. **Challenges**
- For most complex systems, the control equation \( f \) is unknown and the underlying rules have not been fully elucidated. This limits the exploration of these complex fields.
- Although data - driven techniques are helpful in discovering system change patterns, in high - dimensional network dynamics, inferring control equations is still challenging because the number of nodes is usually large, resulting in too many free variables and increasing the complexity of inferring symbolic models.
### Solution
To address these challenges, the authors have developed a general - purpose neuro - symbolic regression tool, called LLC (Learning Law of Changes), which aims to automatically, efficiently and accurately learn the change patterns of complex system states. The main contributions of LLC include:
1. **Decomposing Network Dynamic Signals**
- Introduce physical priors to decompose the control equation \( f \) into two coupled components: auto - dynamics \( Q_{\text{self}} \) and interaction dynamics \( Q_{\text{inter}} \). This can reduce the dimension of high - dimensional network dynamic signals.
- Use neural networks \( \hat{Q}_{\text{self}}^{\theta_1} \) and \( \hat{Q}_{\text{inter}}^{\theta_2} \) to fit these dynamic signals and learn the parameters \( \theta_1 \) and \( \theta_2 \) through back - propagation.
2. **Symbolic Regression to Analyze Control Equations**
- Use a pre - trained symbolic regression model (such as NSRA) to quickly analyze the approximate white - box equations of neural networks, thereby obtaining readable control equations.
3. **Constructing an Executable Pipeline**
- Obtain observational data from initial experiments, select effective time - interval data for difference calculation, then combine physical priors and neural networks to decouple network dynamic signals, and finally analyze the control equations through symbolic regression.
### Experimental Verification
The authors have carried out extensive experimental verifications in multiple fields (such as physics, biochemistry, ecology, epidemiology, etc.), and the results show that LLC has excellent efficiency and accuracy in dealing with complex network dynamics. In particular, in the face of noisy data and topologically missing data, LLC can still maintain a high prediction accuracy.
In conclusion, this paper aims to solve the difficult problem of automatically discovering control equations from complex network dynamics by developing the LLC tool, providing a new general - purpose solution for revealing the hidden mechanisms behind complex phenomena.