The National Science Foundation awarded a $539,926 project grant under the Computer and Information Science and Engineering federal grant program (CFDA 47.070) to the Georgia Tech Research Corporation from September 1, 2022 to August 31, 2025. The grant will support the development of new computational frameworks to improve the generalization of deep transfer learning and reduce model sizes when applying massive pre-trained deep learning models to domains with limited labeled data, such as biomedical applications. Specifically, the grant recipient will develop new adversarial regularization and self-training methods to prevent overfitting and account for noisy labels in limited training data sets. New optimization methods will also be developed to improve training of compact deep learning models in transfer learning scenarios. The recipient will further develop generalization and approximation theories to understand the benefits of the proposed methods. Open-source software libraries facilitating application of these techniques will also be delivered to researchers and practitioners.
Generated 1/6/24, 2:19 PM