<p>This study presents a sample-efficient finite element model updating method tailored for the dynamic analysis of CubeSat structures. To overcome the limitations of conventional machine learning-based approaches, such as excessive sampling and computational inefficiency, two optimization strategies are proposed: one leveraging deep neural networks with continual learning and sensitivity analysis, and the other based on Bayesian optimization incorporating uncertainty-aware sampling. These methods aim to enhance prediction accuracy while minimizing computational cost. The proposed framework is experimentally validated using ground vibration test data from a real CubeSat, demonstrating its effectiveness in correcting modeling errors and improving the fidelity of dynamic response predictions. Notably, the Bayesian approach achieves reliable model updates with significantly fewer simulations, highlighting its practical advantage in high-dimensional design spaces. This research provides a generalizable and efficient finite element model updating strategy that can be readily applied to structural verification tasks in aerospace and other engineering domains. </p>