Using the SEKF to Transfer NN Models of Dynamical Systems with Limited Data
This addresses data scarcity in practical applications like damped spring and reactor systems, but it is incremental as it builds on existing transfer learning and filtering methods.
The paper tackles the problem of adapting pre-trained neural network models to new dynamical systems with limited data by using the Subset Extended Kalman Filter (SEKF), achieving results that require as little as 1% of original training data while reducing computational cost and generalization error.
Data-driven models of dynamical systems require extensive amounts of training data. For many practical applications, gathering sufficient data is not feasible due to cost or safety concerns. This work uses the Subset Extended Kalman Filter (SEKF) to adapt pre-trained neural network models to new, similar systems with limited data available. Experimental validation across damped spring and continuous stirred-tank reactor systems demonstrates that small parameter perturbations to the initial model capture target system dynamics while requiring as little as 1% of original training data. In addition, finetuning requires less computational cost and reduces generalization error.