Randomized selection on the GPU

Monroe, Laura; Michalak, Sarah; Wendelberger, Joanne

doi:10.2172/1090659

Cited by 4 publications

(8 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The main focus of this article is bucketSelect as it outperforms the other algorithms for nonadversarial data and is the most novel of our contributions. Finally, we compare our algorithm to recent related work by Beliakov [2011] and Monroe et al [2011]. The software package, Grinnell GPU k-Selection, contains all of these algorithms with the testing suite and is available online [Alabi et al 2011].…”

Section: Organizationmentioning

confidence: 99%

“…In March 2011, Beliakov [2011] introduced a selection algorithm which chooses pivots based on convex optimization. In August 2011, Monroe et al [2011] introduced a deterministic algorithm based on a random selection of pivots. Both of these algorithms outperform sort&choose and have their own advantages and disadvantages.…”

Section: Related Work: Cuttingplane and Randomizedselectmentioning

confidence: 99%

“…This is the approach taken by Monroe et al [2011] in the algorithm we refer to as randomizedSelect. In this algorithm, a random subset of elements from the vector are chosen as endpoints for assignment bins.…”

Section: Randomizedselectmentioning

confidence: 99%

“…The elements of the vector are then assigned to these three bins, and if the kth-largest element lies in the middle bin, the middle bin is sorted. In Monroe et al [2011], the authors performed testing on unsigned integers with p k = .8. In our tests, we found that using p k = .9 provided faster mean times on floats and doubles.…”

Section: Randomizedselectmentioning

confidence: 99%

“…Our implementation of randomizedSelect is based on Monroe et al [2011] and personal correspondence with Monroe. In addition to the difference in setting p k = .9, there are two other differences to the implementation by Monroe et al First, they use CUDPP sort; this sorting technique is not capable of sorting doubles and the most recent release of CUDPP has adopted the radix sort of Merrill and Grimshaw [CUDPP.org 2011].…”

Section: Randomizedselectmentioning

confidence: 99%

See 4 more Smart Citations

Fast <it>k</it>-selection algorithms for graphics processing units

Alabi

Blanchard

Gordon

et al. 2012

ACM J. Exp. Algorithmics

View full text Add to dashboard Cite

Finding the kth-largest value in a list of n values is a well-studied problem for which many algorithms have been proposed. A naïve approach is to sort the list and then simply select the kth term in the sorted list. However, when the sorted list is not needed, this method does quite a bit of unnecessary work. Although sorting can be accomplished efficiently when working with a graphics processing unit (GPU), this article proposes two GPU algorithms, radixSelect and bucketSelect, which are several times faster than sorting the vector. As the problem size grows so does the time savings of these algorithms with a sixfold speed-up over GPU sorting for float vectors larger than 2 24 and for double vectors larger than 2 20 , ultimately reaching a 19.1 times speed-up for double vectors of length 2 28 .

show abstract

Section: Organizationmentioning

confidence: 99%

Section: Related Work: Cuttingplane and Randomizedselectmentioning

confidence: 99%

Section: Randomizedselectmentioning

confidence: 99%

Section: Randomizedselectmentioning

confidence: 99%

Section: Randomizedselectmentioning

confidence: 99%

See 3 more Smart Citations

Fast <it>k</it>-selection algorithms for graphics processing units

Alabi

Blanchard

Gordon

et al. 2012

ACM J. Exp. Algorithmics

View full text Add to dashboard Cite

show abstract

Fast scalable selection algorithms for large scale data

Thompson

Miranker

2013

2013 IEEE International Conference on Big Data

View full text Add to dashboard Cite

Selection finding, and its most common form median finding, are used as a measure of central tendency for problems in biology, databases, and graphics. These problems often require selection finding as a subcomponent where it can be called many times, and as such speed is important. The Map/Reduce framework has been shown to be an important tool for creating scalable applications. There are a number of valid implementations of the selection algorithms inside of a Map/Reduce framework, certain of which are compared in this paper. However, as the volume of data increases, subtle theoretical algorithmic implementation differences can lead to significant differences in practical application. Therefore, an efficient and scalable selection finding method has the potential to provide general benefit to a number of applications. This paper compares algorithms that have been redesigned or created for the Map/Reduce framework for the purpose of selection finding, or, finding the k-th ranked element in an unordered set. This paper takes the concepts used from two existing selection algorithms and translates them into a novel method using the Map/Reduce framework with two variations. Each approach uses a different methodology to reduce the total amount of workload needed for a selection. All the algorithms are compared together for scalability and efficiency in a computing cluster environment with up to 256 processing cores. The results show that the methods proposed in this paper outperform several common alternatives in identifying medians with Hadoop, including using sorting, Pig, and BinMedian methods. Our implementations are also available upon request.

show abstract

Billion-Scale Similarity Search with GPUs

Johnson

Douze²,

Jeǵou³

2021

IEEE Trans. Big Data

1,993

1,110

View full text Add to dashboard Cite

Similarity search finds application in specialized database systems handling complex data such as images or videos, which are typically represented by high-dimensional features and require specific indexing structures. This paper tackles the problem of better utilizing GPUs for this task. While GPUs excel at data-parallel tasks, prior approaches are bottlenecked by algorithms that expose less parallelism, such as k-min selection, or make poor use of the memory hierarchy.We propose a design for k-selection that operates at up to 55% of theoretical peak performance, enabling a nearest neighbor implementation that is 8.5× faster than prior GPU state of the art. We apply it in different similarity search scenarios, by proposing optimized design for brute-force, approximate and compressed-domain search based on product quantization. In all these setups, we outperform the state of the art by large margins. Our implementation enables the construction of a high accuracy k-NN graph on 95 million images from the Yfcc100M dataset in 35 minutes, and of a graph connecting 1 billion vectors in less than 12 hours on 4 Maxwell Titan X GPUs. We have open-sourced our approach 1 for the sake of comparison and reproducibility.

show abstract

Randomized selection on the GPU

Cited by 4 publications

References 0 publications

Fast <it>k</it>-selection algorithms for graphics processing units

Fast <it>k</it>-selection algorithms for graphics processing units

Fast scalable selection algorithms for large scale data

Billion-Scale Similarity Search with GPUs

Contact Info

Product

Resources

About