A1559
Title: Privacy-aware data analytics via deconvolution
Authors: Farhad Farokhi - The University of Melbourne (Australia) [presenting]
Abstract: Stochastic programs provide a framework for modelling optimization problems that involve uncertainty, such as machine learning, parameter estimation, and data-driven optimization. Most often the distribution of the uncertainty is not known, and samples must be used to approximate the distribution. In many situations, the distribution must be inferred from noisy data samples, such as privacy-preserving or corrupt samples. This can be done via density deconvolution or distributionally robust optimization. The development of data analytics algorithms that work efficiently under differential privacy is discussed through these approaches.
