LeARN: Learnable and Adaptive Representations for Nonlinear Dynamics in System Identification
Researchers have developed LeARN, a new framework for system identification that uses machine learning to discover mathematical models of dynamical systems. Unlike previous methods like SINDy, LeARN learns the necessary basis functions directly from data, eliminating the need for domain-specific expertise. The framework employs meta-learning with a deep neural network to adapt these basis functions, enabling it to handle varying noise conditions and new dynamical regimes effectively. LeARN demonstrates competitive performance on the Neural Fly dataset, marking a step towards more autonomous discovery of complex system principles. AI
IMPACT This framework could accelerate the discovery of governing principles in complex systems by reducing reliance on domain expertise.