Recent Developments of Resampling Methods for Complex Data

In contemporary data science, data structures are becoming increasingly complex due to various factors such as temporal and spatial dependence, non-Euclidean geometry, and high/infinite dimensionality. These complexities pose significant challenges for statistical and predictive inference. As a result, resampling methods such as bootstrap, empirical likelihood, and subsampling have become vital for drawing reliable inferential/predictive results from such complex data. However, existing resampling methods for data with simpler structures often fall short in these complex settings. Naive implementations may fail to capture complex structure within the data, yielding unsatisfactory inferential/predictive results. The goals of this session are to (i) introduce novel resampling methodologies designed to overcome these obstacles and (ii) highlight their benefits to enhancing inference/prediction performance therein. The session will provide chances to share resampling ideas from different settings, and it is therefore expected to inspire new directions for future research for complex data in Statistics and Machine Learning.