CN104360985A - Method and device for implementing clustering algorithm based on MIC - Google Patents
Method and device for implementing clustering algorithm based on MIC Download PDFInfo
- Publication number
- CN104360985A CN104360985A CN201410555944.5A CN201410555944A CN104360985A CN 104360985 A CN104360985 A CN 104360985A CN 201410555944 A CN201410555944 A CN 201410555944A CN 104360985 A CN104360985 A CN 104360985A
- Authority
- CN
- China
- Prior art keywords
- mic
- array
- cpu
- matrix
- class
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Devices For Executing Special Programs (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a method and a device for implementing clustering algorithm based on MIC. The method specifically comprises the following steps: dividing definite MIC arrays and MIC count arrays into one or more than one matrixes; performing matrix calculation on the divided matrixes in a matrix multiplication mode; counting the MIC matrix calculation result, and when the number of changed counts in the MIC matrix calculation result is greater than or equal to a preset threshold, updating the MIC arrays according to the MIC matrix calculation result till the clustering is completed. The device structurally comprises a receiving unit, a dividing unit, a calculating unit, a counting processing unit and a confirming unit. Compared with the prior art, the method and the device for implementing the clustering algorithm based on MIC is used for improving the calculation property and is high in practicability as an MIC coprocessor is adopted.
Description
Technical field
The present invention relates to technical field of data processing, specifically a kind of practical, method and device of realizing clustering algorithm based on MIC.
Background technology
Cluster analysis is also known as cluster analysis, and it is a kind of statistical analysis technique of research (sample or index) classification problem, is also an important algorithm of data mining simultaneously.Cluster analysis, based on similarity, has more similarity between the pattern between the pattern in a cluster than not in same cluster.
K-means algorithm is the hard clustering algorithm of one in cluster analysis, typically based on the objective function clustering method of prototype, K-means algorithm certain distance using data point to prototype, as the objective function optimized, utilizes function to ask the method for extreme value to obtain the regulation rule of interative computation.K-means algorithm, using Euclidean distance as similarity measure, is ask corresponding a certain initial cluster center vector optimal classification, makes evaluation index minimum; Adopt error sum of squares criterion function as clustering criteria function.
Integrated many nuclear technology MIC(Many Integrated Core) be the concurrent coprocessor of the height based on the x86 framework framework that Intel Company will release in the end of the year 2012.Its product line name is called Intel Xeon Phi.
But also do not have a kind of method now, can use MIC optimized algorithm, enable efficient k-means algorithm, more the k-means algorithm of reason random scale is punished in leisure.
Based on this, now provide a kind of method and the device that realize clustering algorithm based on MIC, the method and device realize clustering algorithm based on integrated many nuclear technology MIC, practical.
Summary of the invention
Technical assignment of the present invention is for above weak point, provides a kind of practical, method and device of realizing clustering algorithm based on MIC.
Realize a method for clustering algorithm based on MIC, its specific implementation process is:
One, open up CPU end memory, and MIC array is set according to this CPU end memory;
Two, CPU internal memory is copied to MIC array;
Three, the MIC array determined is divided into one or more 16 matrixes taking advantage of the size of 16; To the matrix taking advantage of 16 less than 16, arrange two zone bits of homography ranks, zone bit is 16bit, wherein row or column in each bit homography, and as there is this row or column, then this bit assignment is 1, otherwise assignment is 0;
Four, to the matrix of each division, carry out matrix operation according to the form of matrix multiplication, to the matrix taking advantage of 16 less than 16, when result adds up, service marking position participates in computing, to obtain coprocessor MIC matrix operation results;
Five, add up MIC matrix operation results, when counting of changing in MIC matrix operation results is more than or equal to the threshold value pre-set, upgrade MIC class array, until cluster completes according to MIC matrix operation results.
CPU end memory in described step one comprises CPU point array and CPU class array, corresponding, and MIC array comprises MIC point array, MIC class array, and its concrete process of opening up is:
From input file, obtain the often scale-of-two of row one group of floating number or the information of textual form, according to the scale-of-two of acquisition or the information of textual form, set up the corresponding CPU point array of central processor CPU internal memory and CPU class array;
According to CPU point array and CPU class array, determine MIC point array, MIC class array.
Described MIC point array, MIC class array deterministic process are: by assignment after the change of CPU point array transpose to MIC point array, by CPU class array indirect assignment to MIC class array.
Initialized step is carried out: the cluster result array that element number is identical with MIC point array is set, and each element of initialization cluster result array is-1 before described matrix operation.
Add up MIC matrix operation results in described step 5 to be realized by cluster result array.
In described step 5, cluster completes the threshold value being less than and pre-setting of counting referring to and change in MIC matrix operation results.
The detailed process upgrading MIC class array according to MIC matrix operation results in described step 5 is: upgrade CPU class array according to MIC matrix operation results, more direct by CPU class array assignment to MIC class array.
Realize a device for clustering algorithm based on MIC, this device comprises receiving element, division unit, computing unit, statistical treatment unit and determining unit; Wherein,
Receiving element, for receiving some array and the class array of CPU process;
Division unit, for being divided into one or more 16 matrixes taking advantage of the size of 16 by the MIC determined class array and MIC point array; To the matrix taking advantage of 16 less than 16, zone bit assignment is 0, and other assignment except zone bit are 1;
Computing unit, for the matrix to each division, carries out matrix operation according to the form of matrix multiplication, and to the matrix taking advantage of 16 less than 16, when result adds up, service marking position participates in computing, to obtain coprocessor MIC matrix operation results;
Statistical treatment unit, for adding up MIC matrix operation results, when counting of changing in MIC matrix operation results is more than or equal to the threshold value pre-set, upgrades MIC class array, until cluster completes according to MIC matrix operation results;
Determining unit, for CPU point array transpose being changed rear assignment to MIC point array, by CPU class array indirect assignment to MIC class array.
A kind of method realizing clustering algorithm based on MIC of the present invention, has the following advantages:
This invention a kind of realizes the method for clustering algorithm based on MIC and device employs MIC coprocessor, improves operational performance; Achieve the cluster to irregular matrix, make K-means clustering method achieve Effec-tive Function under MIC; Practical, applied widely, be easy to promote.
Accompanying drawing explanation
Accompanying drawing 1 realizes the process flow diagram of the method for clustering algorithm for the present invention is based on MIC.
Accompanying drawing 2 realizes the device schematic diagram of clustering algorithm for the present invention is based on MIC.
Embodiment
Below in conjunction with the drawings and specific embodiments, the invention will be further described.
The invention provides a kind of method realizing clustering algorithm based on MIC, as shown in Figure 1, its specific implementation process is:
One, open up CPU end memory, and MIC array is set according to this CPU end memory.
CPU end memory in described step one comprises CPU point array and CPU class array, corresponding, and MIC array comprises MIC point array, MIC class array, and its concrete process of opening up is:
From input file, obtain the often scale-of-two of row one group of floating number or the information of textual form, according to the scale-of-two of acquisition or the information of textual form, set up the corresponding CPU point array of central processor CPU internal memory and CPU class array.
According to CPU point array and CPU class array, determine MIC point array, MIC class array.
Described MIC point array, MIC class array deterministic process are: by assignment after the change of CPU point array transpose to MIC point array, by CPU class array indirect assignment to MIC class array.
Two, CPU internal memory is copied to MIC array.
Three, the MIC array determined is divided into one or more 16 matrixes taking advantage of the size of 16; To the matrix taking advantage of 16 less than 16, arrange two zone bits of homography ranks, zone bit is 16bit, wherein row or column in each bit homography, and as there is this row or column, then this bit assignment is 1, otherwise assignment is 0.
In this step, selecting 16 for calculating dimension, being determined by MIC ardware feature, namely because the major calculations unit of MIC is the vectorization unit of 512 bit wides, 512 bit wides i.e. corresponding 16 floating numbers.Service marking position participates in computing and refers to, uses MIC special purpose function, when performing sum function, being imported into by zone bit, can calculate result as parameter, and namely zone bit is 0 do not add up, zone bit be 1 carrying out add up.
In addition, MIC class array and MIC point array are the arrays completed in MIC internal memory, when only realizing cluster on MIC, just need to determine MIC class array and MIC point array.
Four, to the matrix of each division, carry out matrix operation according to the form of matrix multiplication, to the matrix taking advantage of 16 less than 16, when result adds up, service marking position participates in computing, to obtain coprocessor MIC matrix operation results.
Here, MIC class array and MIC point array are divided into the matrix that 16 take advantage of 16, in the matrix of MIC class array, the element of row can be multiplied according to matrix multiplication with the element that MIC point array arranges, and is similar to the form of matrix multiplication here, refers to this part multiplication portion to replace with to ask poor square.
Below with matrix
be multiplied by
for example, according to the method for matrix multiple, its computing formula is:
.
And according to the inventive method, need each matrix element multiplication portion to replace with and ask poor square, ae, bg, af, bh of being multiplied for above matrix element, need by:
Ae, bg, af, bh replace with respectively: (a-e)
2, (b-g)
2, (a-f)
2, (b-h)
2
It should be noted that, it will be appreciated by those skilled in the art that and be kept in interim array by MIC matrix operation results, the definition of interim array and size etc. belong to the common practise of those skilled in the art.
By to the matrix taking advantage of 16 less than 16, zone bit assignment is 0, and other assignment except zone bit are 1, generates the matrix of rule; When carrying out matrix operation, to the matrix taking advantage of 16 less than 16, when result adds up, service marking position participates in computing, has carried out reduction treatment to assignment section, obtains MIC matrix operation results.
Five, add up MIC matrix operation results, when counting of changing in MIC matrix operation results is more than or equal to the threshold value pre-set, upgrade MIC class array, until cluster completes according to MIC matrix operation results.
Initialized step is carried out: the cluster result array that element number is identical with MIC point array is set, and each element of initialization cluster result array is-1 before described matrix operation.In this step, the content of array should be the call number of other array, therefore value be more than or equal to 0 integer, cluster result array is initialized as negative integer here, in order to avoid initialization value and normal value are obscured, is initialized as-1 traditionally; Why carrying out initialization statement, is to make scheme state sufficiently clear, complete.
Add up MIC matrix operation results in described step 5 to be realized by cluster result array.
In described step 5, cluster completes the threshold value being less than and pre-setting of counting referring to and change in MIC matrix operation results.
The detailed process upgrading MIC class array according to MIC matrix operation results in described step 5 is: upgrade CPU class array according to MIC matrix operation results, more direct by CPU class array assignment to MIC class array.
In above-mentioned steps, upgrade MIC class array according to operation result and refer to according to the numerical value of array mid point array calculating mean value all kinds of in operation result as the class array upgraded how to realize upgrading the conventional techniques means belonging to those skilled in the art.
The present invention is by dividing the MIC class array determined and MIC point array is 16 matrixes taking advantage of the size of 16; To the matrix taking advantage of 16 less than 16, zone bit assignment is 0, and other assignment except zone bit are 1; To the matrix taking advantage of 16 less than 16, when result adds up, service marking position participates in computing, to obtain MIC matrix operation results.Achieve the cluster to irregular matrix, make K-means clustering method achieve Effec-tive Function under MIC.
Realize a device for clustering algorithm based on MIC, as shown in Figure 2, this device comprises receiving element, division unit, computing unit, statistical treatment unit and determining unit; Wherein,
Receiving element, for receiving some array and the class array of CPU process;
Division unit, for being divided into one or more 16 matrixes taking advantage of the size of 16 by the MIC determined class array and MIC point array; To the matrix taking advantage of 16 less than 16, zone bit assignment is 0, and other assignment except zone bit are 1;
Computing unit, for the matrix to each division, carries out matrix operation according to the form of matrix multiplication, and to the matrix taking advantage of 16 less than 16, when result adds up, service marking position participates in computing, to obtain coprocessor MIC matrix operation results;
Statistical treatment unit, the i.e. dotted line frame of upside in accompanying drawing 2, for adding up MIC matrix operation results, when counting of changing in MIC matrix operation results is more than or equal to the threshold value pre-set, MIC class array is upgraded, until cluster completes according to MIC matrix operation results;
Determining unit, the dotted line frame on the downside of namely in accompanying drawing 2, for CPU point array transpose being changed rear assignment to MIC point array, by CPU class array indirect assignment to MIC class array.
Embodiment:
Suppose that existence one includes the input file of the some array of 100 points, the dimension of array file is 18(and represents with 18 numbers for each o'clock), need these arrays to be divided into 10 classes.
First, in CPU, create the two-dimensional array of 100*18 as CPU point array, and the two-dimensional array creating 10*18 is as CPU class array, reads CPU point array and CPU class array according to the conventional process of those skilled in the art from input file.
Equally, in MIC internal memory, set up MIC point array and MIC class array, in order to clearly demonstrate the present embodiment, counting according to MIC accordingly and setting up vertical cluster result array and the interim array for preserving MIC matrix operation results.Wherein ,-1 is initialized as to each element of cluster result array.Here, MIC point array is that CPU point array transpose obtains, and therefore when carrying out matrix computations, needs the definition carrying out being correlated with.
By assignment after CPU point array transpose to MIC point array, by CPU class array indirect assignment to MIC class array.Dividing the MIC class array determined and MIC point array is 16 take advantage of the matrix of the size of 16 (employing shared storage processes); To the matrix taking advantage of 16 less than 16, zone bit assignment is 0, and other assignment except zone bit are 1.
To the matrix of each division, carry out matrix operation according to matrix multiple mode, replaced with by the matrix element be multiplied subtract each other and to ask square in matrix operation by being multiplied, (hypothesis matrix is multiplied middle matrix of consequence element c=a*b, then the inventive method changes c=(a-b) into
2.) other complete according to the mode of being multiplied, and to the matrix taking advantage of 16 less than 16, when result adds up, are multiplied by zone bit, to obtain MIC matrix operation results.Here, 16 take advantage of each element of the matrix of 16 to be that element as a whole carries out processing, the inside multiplication portion of element as a whole, do not adopt replacement to subtract each other and ask square to calculate.
Statistics MIC matrix operation results, when counting of changing in MIC matrix operation results is more than or equal to threshold value, upgrades MIC class array, until cluster completes according to MIC matrix operation results.
Above-mentioned embodiment is only concrete case of the present invention; scope of patent protection of the present invention includes but not limited to above-mentioned embodiment; any according to the invention a kind of based on MIC realize the method for clustering algorithm and claims of device and any person of an ordinary skill in the technical field to its suitable change done or replacement, all should fall into scope of patent protection of the present invention.
Claims (8)
1. realize a method for clustering algorithm based on MIC, it is characterized in that its specific implementation process is:
One, open up CPU end memory, and MIC array is set according to this CPU end memory;
Two, CPU internal memory is copied to MIC array;
Three, the MIC array determined is divided into one or more 16 matrixes taking advantage of the size of 16; To the matrix taking advantage of 16 less than 16, arrange two zone bits of homography ranks, zone bit is 16bit, wherein row or column in each bit homography, and as there is this row or column, then this bit assignment is 1, otherwise assignment is 0;
Four, to the matrix of each division, carry out matrix operation according to the form of matrix multiplication, to the matrix taking advantage of 16 less than 16, when result adds up, service marking position participates in computing, to obtain coprocessor MIC matrix operation results;
Five, add up MIC matrix operation results, when counting of changing in MIC matrix operation results is more than or equal to the threshold value pre-set, upgrade MIC class array, until cluster completes according to MIC matrix operation results.
2. a kind of method realizing clustering algorithm based on MIC according to claim 1, it is characterized in that: the CPU end memory in described step one comprises CPU point array and CPU class array, corresponding, MIC array comprises MIC point array, MIC class array, and its concrete process of opening up is:
From input file, obtain the often scale-of-two of row one group of floating number or the information of textual form, according to the scale-of-two of acquisition or the information of textual form, set up the corresponding CPU point array of central processor CPU internal memory and CPU class array;
According to CPU point array and CPU class array, determine MIC point array, MIC class array.
3. a kind of method realizing clustering algorithm based on MIC according to claim 2, it is characterized in that: described MIC point array, MIC class array deterministic process are: by assignment after the change of CPU point array transpose to MIC point array, by CPU class array indirect assignment to MIC class array.
4. according to a kind of described method realizing clustering algorithm based on MIC arbitrary in claims 1 to 3, it is characterized in that: before described matrix operation, carry out initialized step: the cluster result array that element number is identical with MIC point array is set, and each element of initialization cluster result array is-1.
5. a kind of method realizing clustering algorithm based on MIC according to claim 4, be is characterized in that: add up MIC matrix operation results in described step 5 and realized by cluster result array.
6. a kind of method realizing clustering algorithm based on MIC according to claim 4, it is characterized in that: the detailed process upgrading MIC class array according to MIC matrix operation results in described step 5 is: upgrade CPU class array according to MIC matrix operation results, more direct by CPU class array assignment to MIC class array.
7. a kind of method realizing clustering algorithm based on MIC according to claim 4, is characterized in that: in described step 5, cluster completes the threshold value being less than and pre-setting of counting referring to and change in MIC matrix operation results.
8. realize a device for clustering algorithm based on MIC, it is characterized in that: this device comprises receiving element, division unit, computing unit, statistical treatment unit and determining unit; Wherein,
Receiving element, for receiving some array and the class array of CPU process;
Division unit, for being divided into one or more 16 matrixes taking advantage of the size of 16 by the MIC determined class array and MIC point array; To the matrix taking advantage of 16 less than 16, zone bit assignment is 0, and other assignment except zone bit are 1;
Computing unit, for the matrix to each division, carries out matrix operation according to the form of matrix multiplication, and to the matrix taking advantage of 16 less than 16, when result adds up, service marking position participates in computing, to obtain coprocessor MIC matrix operation results;
Statistical treatment unit, for adding up MIC matrix operation results, when counting of changing in MIC matrix operation results is more than or equal to the threshold value pre-set, upgrades MIC class array, until cluster completes according to MIC matrix operation results;
Determining unit, for CPU point array transpose being changed rear assignment to MIC point array, by CPU class array indirect assignment to MIC class array.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410555944.5A CN104360985A (en) | 2014-10-20 | 2014-10-20 | Method and device for implementing clustering algorithm based on MIC |
| PCT/CN2015/070240 WO2016061911A1 (en) | 2014-10-20 | 2015-01-06 | Method and device for implementing clustering algorithm based on mic |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410555944.5A CN104360985A (en) | 2014-10-20 | 2014-10-20 | Method and device for implementing clustering algorithm based on MIC |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104360985A true CN104360985A (en) | 2015-02-18 |
Family
ID=52528248
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410555944.5A Pending CN104360985A (en) | 2014-10-20 | 2014-10-20 | Method and device for implementing clustering algorithm based on MIC |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN104360985A (en) |
| WO (1) | WO2016061911A1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108509270A (en) * | 2018-03-08 | 2018-09-07 | 中国科学院软件研究所 | The high performance parallel implementation method of K-means algorithms on a kind of domestic 26010 many-core processor of Shen prestige |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109490611B (en) * | 2018-10-29 | 2021-03-05 | 宁波三星智能电气有限公司 | Time statistical method for embedded equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102141976A (en) * | 2011-01-10 | 2011-08-03 | 中国科学院软件研究所 | Method for storing diagonal data of sparse matrix and SpMV (Sparse Matrix Vector) realization method based on method |
| CN102509107A (en) * | 2011-10-13 | 2012-06-20 | 西北工业大学 | Local and global consistence classifying method based on sparse decomposition of figure 10 |
| CN102663415A (en) * | 2012-03-19 | 2012-09-12 | 华中科技大学 | GPU based data classification method of conditional random field model |
| CN103106183A (en) * | 2013-01-29 | 2013-05-15 | 福建天晴数码有限公司 | Large-scale sparse matrix multiplication method based on mapreduce |
| CN103177414A (en) * | 2013-03-27 | 2013-06-26 | 天津大学 | Structure-based dependency graph node similarity concurrent computation method |
| CN104090940A (en) * | 2014-06-27 | 2014-10-08 | 华中科技大学 | Sequential network and sequential data polymorphic clustering method |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100930799B1 (en) * | 2007-09-17 | 2009-12-09 | 한국전자통신연구원 | Automated Clustering Method and Multipath Clustering Method and Apparatus in Mobile Communication Environment |
| CN101923712A (en) * | 2010-08-03 | 2010-12-22 | 苏州大学 | Particle swarm optimization-based gene chip image segmenting method of K-means clustering algorithm |
| CN103440229B (en) * | 2013-08-12 | 2017-11-10 | 浪潮电子信息产业股份有限公司 | A kind of vectorization optimization method based on MIC architecture processors |
| CN103744935B (en) * | 2013-12-31 | 2017-06-06 | 华北电力大学（保定） | A kind of quick mass data clustering processing method of computer |
| CN104123372B (en) * | 2014-07-25 | 2018-03-09 | 浪潮(北京)电子信息产业有限公司 | A kind of method and device that cluster is realized based on CUDA |
-
2014
- 2014-10-20 CN CN201410555944.5A patent/CN104360985A/en active Pending
-
2015
- 2015-01-06 WO PCT/CN2015/070240 patent/WO2016061911A1/en active Application Filing
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102141976A (en) * | 2011-01-10 | 2011-08-03 | 中国科学院软件研究所 | Method for storing diagonal data of sparse matrix and SpMV (Sparse Matrix Vector) realization method based on method |
| CN102509107A (en) * | 2011-10-13 | 2012-06-20 | 西北工业大学 | Local and global consistence classifying method based on sparse decomposition of figure 10 |
| CN102663415A (en) * | 2012-03-19 | 2012-09-12 | 华中科技大学 | GPU based data classification method of conditional random field model |
| CN103106183A (en) * | 2013-01-29 | 2013-05-15 | 福建天晴数码有限公司 | Large-scale sparse matrix multiplication method based on mapreduce |
| CN103177414A (en) * | 2013-03-27 | 2013-06-26 | 天津大学 | Structure-based dependency graph node similarity concurrent computation method |
| CN104090940A (en) * | 2014-06-27 | 2014-10-08 | 华中科技大学 | Sequential network and sequential data polymorphic clustering method |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108509270A (en) * | 2018-03-08 | 2018-09-07 | 中国科学院软件研究所 | The high performance parallel implementation method of K-means algorithms on a kind of domestic 26010 many-core processor of Shen prestige |
| CN108509270B (en) * | 2018-03-08 | 2020-09-29 | 中国科学院软件研究所 | High-performance parallel implementation method of K-means algorithm on domestic Shenwei 26010 many-core processor |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2016061911A1 (en) | 2016-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3343392B1 (en) | Hardware accelerator architecture and template for web-scale k-means clustering | |
| US9886418B2 (en) | Matrix operands for linear algebra operations | |
| Chen et al. | Flexible hardware architecture of hierarchical K-means clustering for large cluster number | |
| Shaydulin et al. | Relaxation-based coarsening for multilevel hypergraph partitioning | |
| CN112633309A (en) | Efficient query black box anti-attack method based on Bayesian optimization | |
| Kułaga et al. | FPGA implementation of decision trees and tree ensembles for character recognition in Vivado HLS | |
| CN102362274A (en) | Processing of linear systems of equations | |
| Kim et al. | Assessment of the number of components in Gaussian mixture models in the presence of multiple local maximizers | |
| Grechkin et al. | Pathway graphical lasso | |
| CN110083969B (en) | Digital integrated circuit layout method based on discrete optimization and terminal equipment | |
| CN110188825B (en) | Image clustering method, system, device and medium based on discrete multi-view clustering | |
| Fierro et al. | The Hawkes process with different exciting functions and its asymptotic behavior | |
| CN104408019A (en) | Method for realizing GMRES (generalized minimum residual) algorithm parallel acceleration on basis of MIC (many integrated cores) platform | |
| CN104933080B (en) | A kind of method and device of determining abnormal data | |
| Tu et al. | Making fisher discriminant analysis scalable | |
| CN104360985A (en) | Method and device for implementing clustering algorithm based on MIC | |
| Liu et al. | A test for equality of two distributions via jackknife empirical likelihood and characteristic functions | |
| CN104615516B (en) | The method that extensive high-performance Linpack test benchmark towards GPDSP is realized | |
| Song et al. | Poster: A memory-access-efficient adaptive implementation of knn on FPGA through hls | |
| CN109978043B (en) | Target detection method and device | |
| CN104572588A (en) | Matrix inversion processing method and device | |
| CN104573696B (en) | Method and apparatus for handling face characteristic data | |
| US20230072535A1 (en) | Error mitigation for sampling on quantum devices | |
| CN104123372B (en) | A kind of method and device that cluster is realized based on CUDA | |
| US10761847B2 (en) | Linear feedback shift register for a reconfigurable logic unit |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150218 |