A multi-institution team of computer scientists has clarified the first general-purpose method of solving nearest neighbor question for complex data. Massachusetts Institute of Technology's Piotr Indyk, a 2015 ACM Fellow and recipient of the 2012 ACM Paris Kanellakis Theory and Practice Award, says, "This is the first result that captures a rich collection of spaces using a single algorithmic technique."
Key to the breakthrough was the examination of factors that prevent an effective nearest neighbor algorithm from existing for a distance metric. The researchers sought a way to embed an expander metric within other distance metrics to prove that they had unworkable expander-like characteristics. The team's research has reimagined the nearest neighbor search for high-dimensional data in a generalized model, and instead of working up one-off algorithms for specific distances, scientists have a universal strategy for finding nearest neighbor algorithms.
From Quanta Magazine
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found