Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The authors derive finite-sample PAC-Bayesian generalization bounds for partially observed stochastic linear time-invariant (LTI) state-space systems driven by control inputs and sub-Gaussian noise. Establishing these finite-sample guarantees provides theoretical rigor for learning dynamical systems from limited trajectory data where asymptotic analysis falls short. The resulting bounds directly link empirical training loss to expected prediction performance, yielding simultaneous prediction and parameter estimation error guarantees across a wide class of system identification algorithms.
Learning dynamical systems from finite noisy trajectories no longer requires asymptotic assumptions: partially observed LTI state-space models now come with rigorous PAC-Bayesian guarantees for both prediction and parameter recovery.
In this paper we derive a Probably Approximately Correct (PAC)-Bayesian error bound for partially observed linear time-invariant (LTI) stochastic dynamical systems in state-space form with inputs and sub-Gaussian noise. Such bounds are widespread in machine learning, and they are useful for characterizing the predictive power of models learned from finitely many data points. The bound derived in this paper relates the expectation of prediction errors with the prediction error generated by the model on the data used for learning. In addition, we show that it can also be used to derive bounds for the parameter estimation error. In turn, this allows us to provide finite-sample error bounds for the prediction error and parameter estimation error for a wide class of system identification algorithms. Furthermore, as LTI systems are a sub-class of recurrent neural networks (RNNs), these error bounds could be a first step towards PAC-Bayesian bounds for RNNs.