Abstract:System identification, the process of deriving mathematical models of dynamical systems from observed input-output data, has undergone a paradigm shift with the advent of learning-based methods. Addressing the intricate challenges of data-driven discovery in nonlinear dynamical systems, these methods have garnered significant attention. Among them, Sparse Identification of Nonlinear Dynamics (SINDy) has emerged as a transformative approach, distilling complex dynamical behaviors into interpretable linear combinations of basis functions. However, SINDy relies on domain-specific expertise to construct its foundational "library" of basis functions, which limits its adaptability and universality. In this work, we introduce a nonlinear system identification framework called LeARN that transcends the need for prior domain knowledge by learning the library of basis functions directly from data. To enhance adaptability to evolving system dynamics under varying noise conditions, we employ a novel meta-learning-based system identification approach that uses a lightweight deep neural network (DNN) to dynamically refine these basis functions. This not only captures intricate system behaviors but also adapts seamlessly to new dynamical regimes. We validate our framework on the Neural Fly dataset, showcasing its robust adaptation and generalization capabilities. Despite its simplicity, our LeARN achieves competitive dynamical error performance compared to SINDy. This work presents a step toward the autonomous discovery of dynamical systems, paving the way for a future where machine learning uncovers the governing principles of complex systems without requiring extensive domain-specific interventions.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to learn an interpretable basis function library of nonlinear dynamical systems from data without prior domain knowledge, in order to achieve efficient identification and adaptation of complex dynamical systems**. Specifically, traditional system identification methods (such as SINDy) rely on the basis function library constructed by domain experts, which limits their flexibility and universality. To solve this problem, the author proposes a new framework named LeARN. LeARN directly learns the basis function library from data through meta - learning, thereby improving the adaptability and generalization ability of the model, especially when facing different noise conditions and changing dynamics. ### Main problems and challenges 1. **Dependence on domain knowledge**: Traditional methods such as SINDy require domain experts to construct the basis function library, which limits their application in new systems or environments. 2. **Insufficient adaptability**: Existing methods are difficult to adapt to changing dynamics and noise conditions. 3. **Complex nonlinear dynamics**: Many robotic systems have high - dimensional and nonlinear characteristics and are difficult to be accurately described by traditional physical models. 4. **Unknown or unmodeled dynamics**: Environmental perturbations and uncertainties introduce dynamics that are difficult to be captured by predefined equations. 5. **High - dimensional input**: The large - scale data generated by robots with multiple sensors and actuators are difficult to be processed by physical - based models. ### Solutions The LeARN framework solves the above problems in the following ways: - **Directly learn the basis function library from data**: There is no need for a pre - constructed basis function library, which enhances flexibility and universality. - **Meta - learning enhances adaptability**: Use a lightweight deep neural network (DNN) to dynamically optimize the basis function library, enabling it to seamlessly adapt to new dynamic patterns. - **Improve interpretability**: Although a deep - learning method is used, a certain degree of interpretability is still maintained by learning the basis function library. ### Verification and application The author verifies the effectiveness of the LeARN framework on the Neural Fly dataset and shows its robust adaptation and generalization ability under different wind speed conditions. The experimental results show that LeARN can be comparable to SINDy in performance while overcoming the latter's dependence on domain knowledge. In conclusion, LeARN provides a new way for machine learning to autonomously discover the dynamic principles of complex systems without a large amount of domain intervention.

LeARN: Learnable and Adaptive Representations for Nonlinear Dynamics in System Identification

LES-SINDy: Laplace-Enhanced Sparse Identification of Nonlinear Dynamical Systems

SINDy vs Hard Nonlinearities and Hidden Dynamics: a Benchmarking Study

Sparse Identification of Nonlinear Dynamics with Side Information (SINDy-SI)

Deep Generative Modeling for Identification of Noisy, Non-Stationary Dynamical Systems

Rapid Bayesian identification of sparse nonlinear dynamics from scarce and noisy data

Iterative Sparse Identification of Nonlinear Dynamics

A Robust SINDy Approach by Combining Neural Networks and an Integral Form

EKF-SINDy: Empowering the extended Kalman filter with sparse identification of nonlinear dynamics

Adaptive Kernel Leaning Networks with Application to Nonlinear System Identification

Weak-form modified sparse identification of nonlinear dynamics

Learning Nonlinear Dynamics Using Kalman Smoothing

Kernel functions embed into the autoencoder to identify the sparse models of nonlinear dynamics

Sparse identification of nonlinear dynamics in the presence of library and system uncertainty

AI-Lorenz: A physics-data-driven framework for Black-Box and Gray-Box identification of chaotic systems with symbolic regression

Generalizing the SINDy approach with nested neural networks

Learning sparse nonlinear dynamics via mixed-integer optimization

GN-SINDy: Greedy Sampling Neural Network in Sparse Identification of Nonlinear Partial Differential Equations

ADAM-SINDy: An Efficient Optimization Framework for Parameterized Nonlinear Dynamical System Identification

Online learning in bifurcating dynamic systems via SINDy and Kalman filtering

Nonlinear system identification in coherence with nonlinearity measure for dynamic physical systems—case studies