An opportunity for two talented students to undertake their PhDs on two projects that concentrate on developing differentially private approaches and blockchain-based solutions for privacy-preserving trustworthy distributed machine learning.
In this research, the successful candidates will focus on mathematical backgrounds involved in differential privacy to devise novel scalable approaches suitable for the privacy preservation of distributed machine learning scenarios such as federated learning, split learning, and distributed stochastic gradient descent. The existing differentially private machine learning approaches show issues in model performance and efficiency and vulnerability towards the application of noise over a large number of distributed models. These issues should be overcome by developing robust and feasible mathematical models for differential privacy by investigating the data dynamics (IID and Non-IID) of distributed machine learning. Besides, trustworthiness is another major property that needs to be investigated in distributed machine learning. Blockchain-based approaches are gaining much attention due to the inherent data immutability and trustworthiness properties offered by the underlying architectures. However, the high latency introduced by blockchain makes timely machine learning decision-making challenging. Hence, in this research, the students will also conduct thorough research in implementing feasible blockchain-based solutions to support trustworthy machine learning.
