Optimal Discrepancy

Optimal discrepancy research focuses on quantifying and minimizing the irregularity of point distributions or the difference between probability distributions, aiming for more efficient and accurate algorithms across diverse applications. Current research explores efficient algorithms for computing and minimizing discrepancy, including novel approaches using graph neural networks to generate low-discrepancy point sets and methods leveraging maximum discrepancy for sample-efficient model evaluation. These advancements have significant implications for various fields, improving the performance of numerical integration, machine learning model training and evaluation, and addressing challenges in domain adaptation and black-box attacks.