Theory Lunch Seminar

  • ILYA RAZENSHTEYN
  • Senior Researcher
  • Machine Learning and Optimization Group
  • Microsoft Research Redmond
Seminars

Random space partitions and optimal transport: beyond metric embeddings

 I will talk about two new results (both joint with Yihe Dong, Piotr Indyk and Tal Wagner):

  * A new algorithm that finds nearest-neighbor-friendly partitions of a high-dimensional space using graph partitioning coupled with supervised learning (e.g., neural networks) https://arxiv.org/abs/1901.08544
  * A new algorithm for optimal transport based on tree approximations https://arxiv.org/abs/1910.04126

The unifying theme is going beyond low-distortion embeddings into simple metric spaces.

About the Speaker

Theory Group YouTube Channel

For More Information, Please Contact: 
Keywords: