Medical Data Privacy
Development of novel methods to train deep learning models on medical data that are sensitive to the unique challenges of patient privacy.
Distributed Learning
Collecting patient data into a centralized database for training a deep learning model is difficult and fraught with concerns about patient data privacy and security. Distributed learning enables models to be trained without the need to collect private patient data into a centralized repository.
Vulnerability of deep learning techniques to attacks
The purpose of the project is to examine the vulnerability of the different segmentation, and classification tasks to attacks that attempt to reconstruct original patient information.