English

Linear RNNs Provably Learn Linear Dynamic Systems

Machine Learning 2023-10-24 v2 Systems and Control Systems and Control

Abstract

We study the learning ability of linear recurrent neural networks with Gradient Descent. We prove the first theoretical guarantee on linear RNNs to learn any stable linear dynamic system using any a large type of loss functions. For an arbitrary stable linear system with a parameter ρC\rho_C related to the transition matrix CC, we show that despite the non-convexity of the parameter optimization loss if the width of the RNN is large enough (and the required width in hidden layers does not rely on the length of the input sequence), a linear RNN can provably learn any stable linear dynamic system with the sample and time complexity polynomial in 11ρC\frac{1}{1-\rho_C}. Our results provide the first theoretical guarantee to learn a linear RNN and demonstrate how can the recurrent structure help to learn a dynamic system.

Keywords

Cite

@article{arxiv.2211.10582,
  title  = {Linear RNNs Provably Learn Linear Dynamic Systems},
  author = {Lifu Wang and Tianyu Wang and Shengwei Yi and Bo Shen and Bo Hu and Xing Cao},
  journal= {arXiv preprint arXiv:2211.10582},
  year   = {2023}
}

Comments

14 pages

R2 v1 2026-06-28T06:15:32.755Z