SYNTHIA will be presenting at the SaTML 2026, the 4th IEEE Conference on Secure and Trustworthy Machine Learning, taking place 23–25 March 2026 in Munich, Germany.
SYNTHIA partner Bogdan Kulynych, Lausanne University Hospital (CHUV), will present the study: “Gaussian DP for Reporting Differential Privacy Guarantees in Machine Learning”
Bogdat will present his abstract in the SaTML 2026 session titled "Differential Privacy".
- Time: 10:00–11:00 CET
- Date: 24 March 2026
About the abstract: Current practices for reporting the level of differential privacy (DP) protection for machine learning (ML) algorithms such as DP-SGD provide an incomplete and potentially misleading picture of the privacy guarantees. For instance, if only a single (ε,δ) is known about a mechanism, standard analyses show that there exist highly accurate inference attacks against training data records, when, in fact, such accurate attacks might not exist. In this position paper, we argue that using non-asymptotic Gaussian Differential Privacy (GDP) as the primary meansofcommunicating DPguarantees in MLavoidsthese potential downsides. Using two recent developments in the DP literature: (i) open-source numerical accountants capable of computing the privacy profile and f-DP curves of DP-SGD to arbitrary accuracy, and (ii) a decision-theoretic metric over DP representations, we show how to provide non-asymptotic bounds on GDP using numerical accountants, and show that GDP can capture the entire privacy profile of DP-SGD and related algorithms with virtually no error, as quantified by the metric. To support the claims, the research team investigated the privacy profiles of state-of-the-art DP large-scale image classification, and the TopDown algorithm for the U.S. Decennial Census, observing that GDP fits their profiles remarkably well in all cases. The research team concludes with a discussion on the strengths and weaknesses of this approach, and discuss which other privacy mechanisms could benefit from GDP. Link to the abstract at Arxiv Computer Science >
This work contributes to ongoing efforts within SYNTHIA to strengthen privacy-preserving approaches in machine learning, supporting the safe and responsible use of health data.
About SaTML 2026
The SaTML conference is sponsored by the IEEE Computer Society Technical Committee of Security and Privacy and brings together researchers and practitioners working to advance the security, robustness, and trustworthiness of machine learning systems. It aims to deepen both theoretical and practical understanding of vulnerabilities in ML, while fostering a unified scientific community dedicated to trustworthy AI. Topics of interest include Novel attacks on machine learning, Novel defenses for machine learning, Secure and safe machine learning in practice, Verification of algorithms and systems, Privacy in machine learning, Forensic analysis of machine learning, Fairness and interpretability and Trustworthy data curation.
