Privacy Preserving Learning

Privacy-preserving learning aims to enable machine learning model training and deployment without compromising the privacy of sensitive data used for training. Current research focuses on developing algorithms that incorporate differential privacy, federated learning techniques (like FedIT), and novel data encryption methods (e.g., using random orthogonal matrices or "human-imperceptible, machine-recognizable" image transformations) to achieve this goal. These advancements are crucial for responsible AI development, particularly in healthcare and other domains with stringent data privacy regulations, enabling the use of sensitive data for model training while mitigating privacy risks.