Privacy Recovery

Privacy recovery research focuses on mitigating the leakage of sensitive information from various sources, particularly large language models and pre-trained models used in machine learning. Current efforts concentrate on developing techniques to remove or mask private data within inputs before processing, and then restoring necessary information for accurate model outputs, or on reconstructing private information from model parameters themselves using novel algorithms like distribution discrimination. This field is crucial for safeguarding individual privacy in the age of ubiquitous data collection and AI, impacting both the responsible development of AI systems and the design of privacy-preserving data sharing mechanisms.