Leverage Score Sampling

Leverage score sampling is a technique used to efficiently approximate computationally expensive operations in machine learning, particularly those involving high-dimensional data or complex models like those used for Shapley value estimation. Current research focuses on developing improved algorithms for leverage score computation and inversion, exploring their application in various contexts such as active learning, numerical integration, and solving large-scale regression problems, often within a randomized linear algebra framework. This work is significant because it enables the application of powerful machine learning methods to larger datasets and more complex models while maintaining accuracy and reducing computational cost, impacting fields ranging from explainable AI to data analysis and scientific computing.