Property Inference

Property inference investigates how to extract information about a model's training data without direct access to it, focusing on inferring statistical properties rather than individual data points. Current research explores this across various machine learning models, including federated learning systems, graph neural networks, and large language models, employing techniques like analyzing model outputs, manipulating model architectures, and leveraging formal logic. This field is crucial for assessing and mitigating privacy risks in machine learning applications and for understanding the representational capabilities of different model architectures, impacting both the development of privacy-preserving techniques and the trustworthiness of AI systems.