AF: Small: Fundamental High-Dimensional Algorithms

点击翻译按钮获取中文摘要

The availability of high-dimensional data in a myriad of applications makes efficient and general-purpose algorithmic tools a critical need. This project explores basic questions to address this need in high dimension: How to compute the volume? How to optimize with constraints? How to sample from a desired distribution? How to learn from samples? Progress on these questions will build the foundations of the emerging science of data. The investigator routinely collaborates with scientists from other disciplines to identify specific challenges. He will continue to organize collaborative workshops, write up-to-date research surveys informed by the discoveries of this project, as well as a textbook on Continuous Algorithms, and maintain open-source software for state-of-the-art sampling.The major goal of this project is to extend algorithmic techniques in three ways: from Euclidean to non-Euclidean geometries, from convex to non-convex settings and from optimization to sampling problems. Concretely, it will explore algorithms for non-convex optimization and sampling, for robust clustering and learning (in the presence of adversarial noise), for faster sampling by utilizing Riemannian geometries, and for volume computation by leveraging recent progress on the Kannan-Lovasz-Simonovits hyperplane conjecture. It will also explore such a conjecture for manifolds, and the possibility of a deterministic algorithm for polytope volume.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.