A1614
Title: Influence-guided subsampling
Authors: Lin Wang - Purdue University (United States) [presenting]
Abstract: Collecting labels for every point in a large dataset is often impractical due to measurement and budget constraints. A common remedy is to label only a carefully chosen subset of design points. Computationally efficient subsampling methods are presented that select small, informative subsets from large candidate pools. Unlike most existing approaches, which are tailored to low-dimensional settings, these methods explicitly accommodate high-dimensional regimes where the number of relevant predictors can be comparable to, or even exceed, the full sample size. Theoretical guarantees are described and performance is demonstrated through extensive numerical experiments. Across a wide range of scenarios, the methods consistently outperform existing subsampling techniques, resulting in substantial savings in labeling costs.
