Trading Privacy, Bandwidth and Accuracy in Algorithmic Machine Learning. This project aims to investigate the trade-offs between privacy, communication costs and accuracy of results when learning from

Description

Trading Privacy, Bandwidth and Accuracy in Algorithmic Machine Learning. This project aims to investigate the trade-offs between privacy, communication costs and accuracy of results when learning from users' sensitive data. The project intends to design faster and more accurate algorithms for a wide range of machine learning tasks by developing a novel and widely-applicable algorithmic framework. Expected outcomes of this project include new theoretical tools to guide the design of data-driven decision systems and rigorously analyse their performance and privacy guarantees. Privacy of individuals' information in data analytics pipelines is a key societal concern. This project should lead to significant benefits by strengthening privacy in these pipelines while also improving accuracy and cost-efficiency.. Scheme: Discovery Early Career Researcher Award. Field: 4613 - Theory of Computation. Lead: Dr Clément Canonne