eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Leibniz International Proceedings in Informatics
1868-8969
2020-06-12
1:1
1:3
10.4230/LIPIcs.SEA.2020.1
article
Algorithm Engineering for High-Dimensional Similarity Search Problems (Invited Talk)
Aumüller, Martin
1
https://orcid.org/0000-0002-7212-6476
IT University of Copenhagen, Denmark
Similarity search problems in high-dimensional data arise in many areas of computer science such as data bases, image analysis, machine learning, and natural language processing. One of the most prominent problems is finding the k nearest neighbors of a data point q ∈ ℝ^d in a large set of data points S ⊂ ℝ^d, under same distance measure such as Euclidean distance. In contrast to lower dimensional settings, we do not know of worst-case efficient data structures for such search problems in high-dimensional data, i.e., data structures that are faster than a linear scan through the data set. However, there is a rich body of (often heuristic) approaches that solve nearest neighbor search problems much faster than such a scan on many real-world data sets. As a necessity, the term solve means that these approaches give approximate results that are close to the true k-nearest neighbors. In this talk, we survey recent approaches to nearest neighbor search and related problems.
The talk consists of three parts: (1) What makes nearest neighbor search difficult? (2) How do current state-of-the-art algorithms work? (3) What are recent advances regarding similarity search on GPUs, in distributed settings, or in external memory?
https://drops.dagstuhl.de/storage/00lipics/lipics-vol160-sea2020/LIPIcs.SEA.2020.1/LIPIcs.SEA.2020.1.pdf
Nearest neighbor search
Benchmarking