Topics in trustworthy machine learning and AI: Robustness, privacy and data heterogeneity

About the project

About the Project Machine learning and statistical algorithms are now implemented at a large scale in almost every aspect of our society, significantly impacting our daily lives through their performance. Hence, there is a soaring demand for the development of trustworthy procedures. Some directions include Robust and private learning in heterogenous and distributed environments. Building on previous work in this area, we plan to tackle more challenging data types, such as network and tensor data, and to study the effects of data heterogeneity, privacy, and contamination within a unified framework. Differential privacy of sampling algorithms. Sampling algorithms inherently possess privacy properties due to its probabilistic nature. Although such nature has been explored for simple sampling algorithms, there is considerable room for further investigation into more sophisticated sampling schemes, the effects of subsampling, and relaxations of the conditions on the target distribution. The investigation will be primarily theoretical in nature. The best way to assess fit is by scanning some of the references listed below. If you feel that we might be a good fit, please send your CV , transcript , and the title of a paper you have read from the reference list to m.li.15@bham.ac.uk . If you are shortlisted for an interview, you will be asked to prepare a presentation based on the paper you selected from the reference list. Please assume that your application has been unsuccessful if you do not receive a reply from me within one week.

Research areas