PL2AP: Fast parallel cosine similarity search

David C. Anastasiu; George Karypis

doi:10.1145/2833179.2833182

PL2AP: Fast parallel cosine similarity search

David C. Anastasiu, George Karypis

Computer Science and Engineering

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

8 Scopus citations

Abstract

Solving the AllPairs similarity search problem entails finding all pairs of vectors in a high dimensional sparse dataset that have a similarity value higher than a given threshold. The output form this problem is a crucial component in many real-world applications, such as clustering, online advertising, recommender systems, near-duplicate document detection, and query refinement. A number of serial algorithms have been proposed that solve the problem by pruning many of the possible similarity candidates for each query object, after accessing only a few of their non-zero values. The pruning process results in unpredictable memory access patterns that can reduce search efficiency. In this context, we introduce pL2AP, which efficiently solves the AllPairs cosine similarity search problem in a multi-core environment. Our method uses a number of cache-tiling optimizations, combined with fine-grained dynamically balanced parallel tasks, to solve the problem 1.5x- 238x faster than existing parallel baselines on datasets with hundreds of millions of non-zeros.

Original language	English (US)
Title of host publication	Proceedings of the 5th Workshop on Irregular Applications
Subtitle of host publication	Architectures and Algorithms, IA3 2015
Publisher	Association for Computing Machinery, Inc
ISBN (Electronic)	9781450340014
DOIs	https://doi.org/10.1145/2833179.2833182
State	Published - Nov 15 2015
Event	5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015 - Austin, United States Duration: Nov 15 2015 → …

Publication series

Name	Proceedings of the 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015

Other

Other	5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015
Country/Territory	United States
City	Austin
Period	11/15/15 → …

Bibliographical note

Funding Information:
This work was supported in part by NSF (IIS-0905220, OCI- 1048018, CNS-1162405, IIS-1247632, IIP-1414153, IIS-1447788), Army Research Office (W911NF-14-1-0316), Intel Software and Services Group, and the Digital Technology Center at the University of Minnesota. Access to research and computing facilities was provided by the Digital Technology Center and the Minnesota Supercomputing Institute.

Publisher Copyright:
© 2015 ACM.

Keywords

Bounded cosine similarity graph
Cosine similarity
Similarity join
Similarity search

Access

10.1145/2833179.2833182

OpenUrl availability

Full text

Cite this

Anastasiu, D. C., & Karypis, G. (2015). PL2AP: Fast parallel cosine similarity search. In Proceedings of the 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015 Article a8 (Proceedings of the 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015). Association for Computing Machinery, Inc. https://doi.org/10.1145/2833179.2833182

PL2AP: Fast parallel cosine similarity search. / Anastasiu, David C.; Karypis, George.
Proceedings of the 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015. Association for Computing Machinery, Inc, 2015. a8 (Proceedings of the 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Anastasiu, DC & Karypis, G 2015, PL2AP: Fast parallel cosine similarity search. in Proceedings of the 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015., a8, Proceedings of the 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015, Association for Computing Machinery, Inc, 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015, Austin, United States, 11/15/15. https://doi.org/10.1145/2833179.2833182

@inproceedings{a44c28b446d24ad3aa0fd0082e8ff704,

title = "PL2AP: Fast parallel cosine similarity search",

abstract = "Solving the AllPairs similarity search problem entails finding all pairs of vectors in a high dimensional sparse dataset that have a similarity value higher than a given threshold. The output form this problem is a crucial component in many real-world applications, such as clustering, online advertising, recommender systems, near-duplicate document detection, and query refinement. A number of serial algorithms have been proposed that solve the problem by pruning many of the possible similarity candidates for each query object, after accessing only a few of their non-zero values. The pruning process results in unpredictable memory access patterns that can reduce search efficiency. In this context, we introduce pL2AP, which efficiently solves the AllPairs cosine similarity search problem in a multi-core environment. Our method uses a number of cache-tiling optimizations, combined with fine-grained dynamically balanced parallel tasks, to solve the problem 1.5x- 238x faster than existing parallel baselines on datasets with hundreds of millions of non-zeros.",

keywords = "Bounded cosine similarity graph, Cosine similarity, Similarity join, Similarity search",

author = "Anastasiu, {David C.} and George Karypis",

note = "Funding Information: This work was supported in part by NSF (IIS-0905220, OCI- 1048018, CNS-1162405, IIS-1247632, IIP-1414153, IIS-1447788), Army Research Office (W911NF-14-1-0316), Intel Software and Services Group, and the Digital Technology Center at the University of Minnesota. Access to research and computing facilities was provided by the Digital Technology Center and the Minnesota Supercomputing Institute. Publisher Copyright: {\textcopyright} 2015 ACM.; 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015 ; Conference date: 15-11-2015",

year = "2015",

month = nov,

day = "15",

doi = "10.1145/2833179.2833182",

language = "English (US)",

series = "Proceedings of the 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015",

publisher = "Association for Computing Machinery, Inc",

booktitle = "Proceedings of the 5th Workshop on Irregular Applications",

}

TY - GEN

T1 - PL2AP

T2 - 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015

AU - Anastasiu, David C.

AU - Karypis, George

N1 - Funding Information: This work was supported in part by NSF (IIS-0905220, OCI- 1048018, CNS-1162405, IIS-1247632, IIP-1414153, IIS-1447788), Army Research Office (W911NF-14-1-0316), Intel Software and Services Group, and the Digital Technology Center at the University of Minnesota. Access to research and computing facilities was provided by the Digital Technology Center and the Minnesota Supercomputing Institute. Publisher Copyright: © 2015 ACM.

PY - 2015/11/15

Y1 - 2015/11/15

N2 - Solving the AllPairs similarity search problem entails finding all pairs of vectors in a high dimensional sparse dataset that have a similarity value higher than a given threshold. The output form this problem is a crucial component in many real-world applications, such as clustering, online advertising, recommender systems, near-duplicate document detection, and query refinement. A number of serial algorithms have been proposed that solve the problem by pruning many of the possible similarity candidates for each query object, after accessing only a few of their non-zero values. The pruning process results in unpredictable memory access patterns that can reduce search efficiency. In this context, we introduce pL2AP, which efficiently solves the AllPairs cosine similarity search problem in a multi-core environment. Our method uses a number of cache-tiling optimizations, combined with fine-grained dynamically balanced parallel tasks, to solve the problem 1.5x- 238x faster than existing parallel baselines on datasets with hundreds of millions of non-zeros.

AB - Solving the AllPairs similarity search problem entails finding all pairs of vectors in a high dimensional sparse dataset that have a similarity value higher than a given threshold. The output form this problem is a crucial component in many real-world applications, such as clustering, online advertising, recommender systems, near-duplicate document detection, and query refinement. A number of serial algorithms have been proposed that solve the problem by pruning many of the possible similarity candidates for each query object, after accessing only a few of their non-zero values. The pruning process results in unpredictable memory access patterns that can reduce search efficiency. In this context, we introduce pL2AP, which efficiently solves the AllPairs cosine similarity search problem in a multi-core environment. Our method uses a number of cache-tiling optimizations, combined with fine-grained dynamically balanced parallel tasks, to solve the problem 1.5x- 238x faster than existing parallel baselines on datasets with hundreds of millions of non-zeros.

KW - Bounded cosine similarity graph

KW - Cosine similarity

KW - Similarity join

KW - Similarity search

UR - http://www.scopus.com/inward/record.url?scp=84966783628&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84966783628&partnerID=8YFLogxK

U2 - 10.1145/2833179.2833182

DO - 10.1145/2833179.2833182

M3 - Conference contribution

AN - SCOPUS:84966783628

T3 - Proceedings of the 5th Workshop on Irregular Applications: Architectures and Algorithms, IA3 2015

BT - Proceedings of the 5th Workshop on Irregular Applications

PB - Association for Computing Machinery, Inc

Y2 - 15 November 2015

ER -

PL2AP: Fast parallel cosine similarity search

Abstract

Publication series

Other

Bibliographical note

Keywords

Access

OpenUrl availability

Other files and links

Fingerprint

Cite this