Advanced Data Privacy: Attacks & Defences - ERP Spain.com - España y América Latina

Sensitive information about individuals can be recovered from different types of data releases, including aggregate statistics or machine learning models. This session will address the privacy risks in publishing analysis results and introduce data privacy techniques to defend against them. Theresa Stadler, Data Scientist at Privitar, will explain differencing and reconstruction attacks on simple summary statistics such as count tables, along with discussing the privacy risks of supervised machine learning. Some of the takeaways of the session include: - Reasons to be concerned about the privacy of training data - The attacks on machine learning models that can occur and what private information about individuals in the training data can be recovered - A simple example of a black-box privacy attack on a classifier, a common machine learning model - An introduction to the differential privacy framework that functions as a privacy-enhancing technology to defend against the attacks introduced