Title: Robustness, local differential privacy and change point analysis

 Abstract:

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. The three topics in the title are broadly related to the issue of trust. At a high level, robustness refers to the performance of a procedure in the presence of outliers and/or heavy-tailed errors, while preserving privacy constrains how much information can be shared from potentially sensitive raw data. Change points, on the other hand, are useful in modelling non-stationary time series data with potential application in detecting abnormal behaviours in data streams.

With an appropriate level of introduction to each topic in the title, I will present my thesis which contains three works at their intersection. The titles of these three papers are: Adversarially robust change point detection, Network change point localisation under local differential privacy, and On robustness and local differential privacy. They are accepted by NeurIPS 2021, NeurIPS 2022 and the Annals of Statistics, respectively.