Valerio Maggio: PPML: Machine Learning on Data you cannot see

Speaker:: Valerio Maggio Track: PyData: Data Handling What if I tell you that you can run a complete ML pipeline on private data, without any anonymisation, nor even accessing the data in the first place And what If I also tell you that you can do that with no disruption to your existing pipeline, nor affecting the overall model performance Well, that wouldn t be entirely true but in this workshop we ll explore the great potential privacypreserving machine learning methods have to run machine learning experiments on data you cannot see. In the first part, we will first explore examples of exploits and vulnerabilities of Deep Learning models trained on anonymised data, whilst in the second part we will go much deeper into PPML methods, training DL on encrypted data, and more. You ll just have to be familiar with PyTorch, and DL basics to attend this workshop Recorded at the PyConDE PyData Berlin 2022 conference, April 1113 2022. More details at the conference page: h