Empirical Privacy

Empirical privacy research focuses on evaluating the actual privacy risks of machine learning models, particularly concerning the leakage of sensitive training data. Current efforts concentrate on developing robust attack methods, such as membership inference attacks and their variations (e.g., range membership inference), and creating comprehensive benchmarks for evaluating both the privacy vulnerabilities and the effectiveness of defense mechanisms across various model architectures, including large language models. This field is crucial for ensuring responsible development and deployment of machine learning systems, informing the design of privacy-preserving techniques, and ultimately shaping data protection regulations and practices.