This $400,000 Project Grant from the National Science Foundation Office of Advanced Cyberinfrastructure, under the Computer and Information Science and Engineering program (CFDA 47.070), will fund the development of smart surrogates to accelerate scientific simulations. Over two years from July 2022 to June 2024, Rutgers, The State University will create a new conceptual and infrastructure framework for developing smart surrogates using deep active learning and Bayesian techniques. This will allow surrogate models to intelligently select optimal training data and dynamically adjust neural network architectures. Rutgers will design an open-source toolkit called ROSE to support concurrent execution of simulations and surrogate training tasks in an adaptive manner. Evaluation will focus on applying smart surrogates to diffusion equations and personalized heart simulations. Outcomes aim to directly enable long-term follow-on scientific research while preparing next-generation researchers at the intersection of related fields.