My research interest falls in the scope of Natural Robustness, which preserves model performance under naturally induced data corruptions or alterations across the machine learning pipeline. The goal of my research is to build a reliable learning system, which can work safely and adaptively in the presence of unexpected data conditions across the entire pipeline. My research is closely related to data-centric machine learning with keywords: annotation, augmentation, imbalance, noise, distillation, out-of-distribution, and privacy. Research topics that I am currently working on include:
Exploring non-conventional data sources.
Auditing and handling data quality;
Stress testing and adapting in extreme data conditions.
Protecting data copyright and privacy;