The Statistical learning and differential privacy workshop took place on 12-13 September 2022 at the University of Bath.
Statistical learning and deep learning techniques have been deployed in many parts of our lives, for example in search engines, online recommendation systems, and AI-assisted healthcare. An important question is how can we perform statistical learning to find general patterns from datasets without revealing data of individual participants? This question has become the key challenge that hinders further applications of statistical learning and deep learning in privacy-sensitive applications. Differential Privacy (DP) is a mathematical framework that can provide theoretical guarantees of privacy, while allowing us to achieve model utility and accuracy for specific applications. Mathematics has been the key for breakthroughs in developing statistical learning with DP. Recently, we have seen exciting developments in compressive learning and dynamical systems for designing and proving statistical learning algorithms with DP guarantees.
This workshop brought together researchers and practitioners from statistical machine learning, deep learning, compressive sensing, dynamical systems and Bayesian machine learning to discuss this development and provided a snapshot of this interdisciplinary research topic to students, mathematicians, computer scientists and the wider community.
This workshop was organised by the Center for Mathematics and Algorithms for Data (MAD) It was sponsored by ART-AI and Maths4DL.