Large data real-time classification with Non-negative Matrix Factorization and Self-Organizing Maps on GPU
Published 1 October 2010
Jan Platoš, Petr Gajdoš
Citations4
Generate an AI Snapshot to get a quick, structured summary of this paper.
Study Snapshot
ObjectiveStudy objective
MethodsResearch methodology
PopulationPopulation studied
Sample sizeSample sizes
OutcomesStudy outcomes here
ResultsStudy results comes here
LimitationsResearch study limitations comes here
A concise AI-generated summary of the paper will appear here once you click Generate AI Snapshot.
TL;DR
This article brings an interesting comparison of two different methods, which were implemented on GPU and help us to detect system intrusions and Non-negative Matrix Factorization and Self-Organizing Maps.
Abstract
This article brings an interesting comparison of two different methods, which were implemented on GPU and help us to detect system intrusions. Generally, both of them can be widely used in the area of information retrieval. The modern trends of parallel computation have a significant influence on performance of implemented methods (Non-negative Matrix Factorization (NMF) and Self-Organizing Maps (SOM)). Both methods were compared on real data.
Keywords
Computer Science
Neural Networks: A Comprehensive Foundation
29,816 Citations1998Simon Haykin
Thorough, well-organized, and completely up to date, this book examines all the important aspects of this emerging technology, including the learning process, back-propagation learning, radial-basis function networks, self-organizing systems, modular networks, temporal processing and neurodynamics, and VLSI implementation of neural networks.
BiometricsClassification and Regression Trees.
23,841 Citations1984Alexander Gordon, Leo Breiman +3 more
NatureLearning the parts of objects by non-negative matrix factorization
14,055 Citations1999Daniel D. Lee, H. Sebastian Seung
An algorithm for non-negative matrix factorization is demonstrated that is able to learn parts of faces and semantic features of text and is in contrast to other methods that learn holistic, not parts-based, representations.
Springer series in information sciencesSelf-Organizing Maps
10,377 Citations1997Teuvo Kohonen
The mathematical preliminaries, background, basic ideas, and implications of the Self-Organising Map algorithm are expounded in a manner which is accessible without prior expert knowledge.
Cambridge University Press eBooksPattern Recognition and Neural Networks
6,468 Citations1996B. D. Ripley
Algorithms for Non-negative Matrix Factorization
5,463 Citations2000Daniel D. Lee, H. Sebastian Seung
Two different multiplicative algorithms for non-negative matrix factorization are analyzed and one algorithm can be shown to minimize the conventional least squares error while the other minimizes the generalized Kullback-Leibler divergence.
A Survey of Dimension Reduction Techniques
1,123 Citations2002Imola K. Fodor
Several dimension reduction methods are described that can be applied to high-dimensional datasets to reduce the dimension of the original data prior to any modeling ofmore » the data.
OpenMP to GPGPU
413 Citations2009Seyong Lee, Seung-Jai Min +1 more
This paper presents a compiler framework for automatic source-to-source translation of standard OpenMP applications into CUDA-based GPGPU applications, and identifies several key transformation techniques, which enable efficient GPU global memory access, to achieve high performance.
Society for Industrial and Applied Mathematics eBooksMatrix Methods in Data Mining and Pattern Recognition
326 Citations2007Lars Eldén
This book discusses vectors and matrices in data mining and pattern recognition, Linear Algebra Concepts and Matrix Decompositions, and Computing Eigenvalues and singular values.
Neural Network and Fuzzy Logic Applications in C/C++
150 Citations1994Stephen T. Welstead
A complete, step-by-step guide to developing fuzzy and neural applications; applications for financial forecasting, modeling of chaotic dynamics, pattern classification, and control of unstable systems; provides a wealth of sample code for all applications; critically evaluates how well each application works; and features a guide to creating a complete object-oriented interactive interface framework through which all applications can be run.
Singular value decomposition on GPU using CUDA
130 Citations2009Sheetal Lahabar, P. J. Narayanan
This paper presents the implementation of singular value decomposition (SVD) of a dense matrix on GPU using the CUDA programming model and shows a speedup of upto 60 over the MATLAB implementation and upto 8 over the Intel MKL implementation on a Intel Dual Core 2.66GHz PC for large matrices.
Computational Methods for Intelligent Information Access
129 Citations1995Michael W. Berry, Susan Dumais +1 more
A survey of the computational requirements for managing LSI-encoded databases as well as current and future applications of LSI is presented, with a promising way to improve users’ access to many kinds of textual materials.
Journal of Computational BiologyParallel GPU Implementation of Iterative PCA Algorithms
118 Citations2009M. Andrecut
An algorithm based on Gram-Schmidt orthogonalization (called GS-PCA) is presented, which eliminates this shortcoming of NIPALS- PCA and the numerical results show that the GPU parallel optimized versions, based on CUBLAS (NVIDIA), are substantially faster than the CPU optimized versions based on CBLas (GNU Scientific Library).
Understanding Complex Datasets
107 Citations2007David B. Skillicorn
IEEE Transactions on Knowledge and Data EngineeringCSVD: clustering and singular value decomposition for approximate similarity search in high-dimensional spaces
76 Citations2003V. Castelli, A. Thomasian +1 more
Experiments with texture vectors extracted from satellite images show that CSVD achieves significantly higher dimensionality reduction than plain SVD for the same normalized mean squared error (NMSE), which translates into a higher efficiency in processing approximate NN queries.
New Journal of PhysicsAccelerated fluctuation analysis by graphic cards and complex pattern formation in financial markets
57 Citations2009Tobias Preis, Peter Virnau +2 more
The compute unified device architecture is an almost conventional programming approach for managing computations on a graphics processing unit (GPU) as a data-parallel computing device and significant pattern-based correlations on short time scales are found.
Parallel ComputingParallelizing the Self-Organizing Feature Map on multiprocessor systems
39 Citations1991Chwan-Hwa Wu, R.E. Hodges +1 more
Designing parallel computers for self organizing maps
33 Citations1991Tomas Nordström
The design of massively parallel computers, especially computers with simple processing elements, used for SOM calculations are studied and it is found that SOM (like many other ANN models) demands very little of a parallel computer.
Computers & GeosciencesA parallel Kohonen algorithm for the classification of large spatial datasets
33 Citations1996Stan Openshaw, Ian Turton
A new algorithm suitable for the classification of large spatial datasets suitable for parallel processing using the Kohonen net is presented and its performance on the Cray T3D parallel supercomputer is demonstrated.
NeurocomputingA parallel growing architecture for self-organizing maps with unsupervised learning
30 Citations2005Iren Valova, Daniel Szer +2 more
This work presents a SOM that processes the whole input in parallel and organizes itself over time, and develops networks that do not reorganize their structure from scratch every time a new set of input vectors is presented, but rather adjust their internal architecture in accordance with previous mappings.
Lecture notes in computer scienceHybrid Feature Selection for Modeling Intrusion Detection Systems
28 Citations2004Srilatha Chebrolu, Ajith Abraham +1 more
Empirical results indicate that significant input feature selection is important to design an IDS that is lightweight, efficient and effective for real world detection systems.
Proceedings - IEEE International Parallel and Distributed Processing SymposiumData access optimizations for highly threaded multi-core CPUs with multiple memory controllers
25 Citations2008Georg Hager, Thomas Zeiser +1 more
Using the new Sun UltraSPARC T2 processor as a prototypical multi-core design, performance patterns in low-level and application benchmarks are analyzed and ways to circumvent bottlenecks by careful data layout and padding are shown.
arXiv (Cornell University)A Study of Parallel Self-Organizing Map
22 Citations1998Li Weigang
The paper proves that Parallel-SOM has the same convergence property as Kohonen's SOM, but the complexity of former is reduced obviously.
Accelerator-Oriented Algorithm Transformation for Temporal Data Mining
18 Citations2009Girish Kumar Patnaik, Sean Ponce +2 more
This work presents a novel implementation of a frequent episode discovery algorithm by revisiting 'in-the-large' issues such as problem decomposition as well as 'in the small' issuessuch as data layouts and memory access patterns.
Matrix Factorization Approach for Feature Deduction and Design of Intrusion Detection Systems
13 Citations2008Václav Snåšel, Jan Platoš +2 more
A novel matrix factorization approach for feature deduction and design of intrusion detection systems is proposed and experiment results indicate that the proposed method is efficient.
On the Implementation of Boolean Matrix Factorization
12 Citations2008Václav Snåšel, Pavel Krömer +2 more
The background of the task as well as genetic algorithm based solver are introduced and results obtained from computer experiments are presented.
Normalization Impact on SVD-Based Iris Recognition
11 Citations2009Pavel Moravec, Petr Gajdoš +2 more
This paper compares the performance of singular value decomposition for the recognition of human irises based for both unmodified irises and their normalized versions.
Evolutionary Approaches to Linear Ordering Problem
10 Citations2008Václav Snáel, Pavel Krömer +1 more
This paper investigates the usage and performance of two variants of genetic algorithms - mutation only genetic algorithms and higher level chromosome genetic algorithms- on the linear ordering problem.
Scaling IDS construction based on Non-negative Matrix factorization using GPU computing
8 Citations2010Jan Platoš, Pavel Krömer +2 more
Improvements to some of the previous approaches using a Non-negative Matrix factorization approach are presented and to improve the performance (detection accuracy) and computational speed (scaling) a GPU implementation is detailed.
Large-Scale Principal Component Analysis on LiveJournal Friends Network
6 Citations2008Miklós Kurucz, András A. Benczúr +1 more
This work extends the applicability of PCA to very large scale social networks by handling the abundance of small size communities that hide the higher level structure and promising similar or better performance than semidefinite relaxation in much shorter running time.
Detecting Insider Attacks Using Non-negative Matrix Factorization
6 Citations2009Jan Platoš, Václav Snåšel +2 more
Comparisons with other established pattern recognition techniques reveal that the Non-negative Matrix Factorization approach could be also an ideal candidate to detect internal threats.