Exploring Deep Neural Network Reuse in Computational Natural Science

The creation and training of deep learning models pose significant technical, financial, and environmental challenges, which can be addressed through the reuse of pre-trained deep neural network models (PTMs), which enables deep learning methodologies with little comparative cost. While previous research has focused on PTM applications in software engineering, our work investigates their growing role in computational Natural Science. Our ongoing work examines and longitudinally measures PTM reuse in computational Natural Science. Our efforts quantify PTM reuse methodologies, identifies trends regarding the reuse of PTM architectures across Natural Science, and provides insights into how PTMs are progressing science.