Module 23: Approximate Nearest Neighbor Problem in High Dimensions with Constraints

Faculty Contact: Subhash Suri

Research Area(s): 

Abstract: While the approximate nearest neighbor (ANN) problem in high dimensions has been well-studied, algorithms which solve this problem, as well as most ANN algorithms, rarely consider additional constraints on potential neighbors (those which restrict the set of viable neighbors). This module explores how additional constraints can be introduced to enrich the queries available for nearest neighbor algorithms. In particular, we aim to couple the relationship between distance in vector space and distance in graph space and to determine how this changes search in high dimensions.

Active Quarters:

  • Winter 2017: Alex Jones and Isaac Mackey