Pre-training Aided Medical Data Analysis Under Privacy Constraints

The School of ITEE is hosting the following PhD confirmation progress seminar:

Pre-training Aided Medical Data Analysis Under Privacy Constraints

Speaker: Yixuan Qiu
Host: Dr Miao Xu

Abstract: Machine learning is being used to combat the shortage of physicians through medical data analysis. However, strict privacy constraints limit data sharing between institutions, making it difficult to develop robust models. To overcome this challenge, the research will propose a Privacy-Preserving Selective Learning framework that pre-trains models on selective data. Preliminary work has been done by assuming data from hospitals with insufficient data and aid providers are identically distributed. The framework selects a desired proxy subset from the collaborator’s dataset in a closed-world setting to improve latent representations and enhance medical data analysis. Future work will investigate a more complex situation where external datasets are heterogeneous including out-of-distribution data. A communication-efficient pre-training-aided framework will also be developed to address large pre-training models’ challenges. The research aims to contribute to medical data analysis by addressing data privacy concerns while mitigating the physician shortage issue.

Speaker Biography: Yixuan Qiu obtained his M.E. in Electrical Engineering from the University of Queensland in 2020. He started his PhD in computer science under the supervision of Dr. Miao Xu and Dr. Weitong Chen in 2022. His research interests include medical data analysis and weakly supervised learning.