Student scores 100% on her MRes!
We take great pleasure in announcing that the paper "The Value of Collaboration in Convex Machine Learning with Differential Privacy" http://arxiv.org/abs/1906.09679 has been accepted by the most prestigious conference in Security and Privacy IEEE Security and Privacy Symposium 2020 https://www.ieee-security.org/TC/SP2020/
Congratulations to Nan Wu who scored 100% in her MRes Research Paper in collaboration with Farhad Farokhi and David Smith from Data61. Supervisor Dali Kaafar
The paper develops differentially private gradient descent algorithms for training Machine learning models on distributed private datasets owned by different entities. We prove that the quality of the trained ML model using DP gradient descent algorithm scales inversely with privacy budgets squared, and the size of the distributed datasets squared, which establishes a trade-off between privacy and utility in privacy-preserving ML;
We also develop a theory that enables to predict the outcome of a potential collaboration among privacy-aware data owners (or data custodians) in terms of the fitness cost of the ML training model prior to executing potentially computationally-expensive ML algorithms on distributed privately-owned datasets.