Parallelism in High-Performance Computing Applications

Slides:



Advertisements
Similar presentations
Chapter 28 – Part II Matrix Operations. Gaussian elimination Gaussian elimination LU factorization LU factorization Gaussian elimination with partial.
Advertisements

Latent Semantic Analysis
Dimensionality reduction. Outline From distances to points : – MultiDimensional Scaling (MDS) Dimensionality Reductions or data projections Random projections.
Dimensionality Reduction PCA -- SVD
Solving Linear Systems (Numerical Recipes, Chap 2)
1cs542g-term High Dimensional Data  So far we’ve considered scalar data values f i (or interpolated/approximated each component of vector values.
Hinrich Schütze and Christina Lioma
CS 584. Review n Systems of equations and finite element methods are related.
Principal Component Analysis
3D Geometry for Computer Graphics
Latent Semantic Indexing via a Semi-discrete Matrix Decomposition.
Lecture 19 Quadratic Shapes and Symmetric Positive Definite Matrices Shang-Hua Teng.
1 Latent Semantic Indexing Jieping Ye Department of Computer Science & Engineering Arizona State University
Dimensional reduction, PCA
Vector Space Information Retrieval Using Concept Projection Presented by Zhiguo Li
Avoiding Communication in Sparse Iterative Solvers Erin Carson Nick Knight CS294, Fall 2011.
Information Retrieval in Text Part III Reference: Michael W. Berry and Murray Browne. Understanding Search Engines: Mathematical Modeling and Text Retrieval.
TFIDF-space  An obvious way to combine TF-IDF: the coordinate of document in axis is given by  General form of consists of three parts: Local weight.
The Terms that You Have to Know! Basis, Linear independent, Orthogonal Column space, Row space, Rank Linear combination Linear transformation Inner product.
Introduction to Information Retrieval Introduction to Information Retrieval Hinrich Schütze and Christina Lioma Lecture 18: Latent Semantic Indexing 1.
Clustering In Large Graphs And Matrices Petros Drineas, Alan Frieze, Ravi Kannan, Santosh Vempala, V. Vinay Presented by Eric Anderson.
Previously Two view geometry: epipolar geometry Stereo vision: 3D reconstruction epipolar lines Baseline O O’ epipolar plane.
3D Geometry for Computer Graphics
Lecture 20 SVD and Its Applications Shang-Hua Teng.
Kathryn Linehan Advisor: Dr. Dianne O’Leary
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. Parallel Programming in C with MPI and OpenMP Michael J. Quinn.
Multimedia Databases LSI and SVD. Text - Detailed outline text problem full text scanning inversion signature files clustering information filtering and.
E.G.M. PetrakisDimensionality Reduction1  Given N vectors in n dims, find the k most important axes to project them  k is user defined (k < n)  Applications:
DATA MINING LECTURE 7 Dimensionality Reduction PCA – SVD
SVD(Singular Value Decomposition) and Its Applications
Summarized by Soo-Jin Kim
CS246 Topic-Based Models. Motivation  Q: For query “car”, will a document with the word “automobile” be returned as a result under the TF-IDF vector.
Next. A Big Thanks Again Prof. Jason Bohland Quantitative Neuroscience Laboratory Boston University.
1 Information Retrieval through Various Approximate Matrix Decompositions Kathryn Linehan Advisor: Dr. Dianne O’Leary.
Non Negative Matrix Factorization
SVD: Singular Value Decomposition
CpSc 881: Information Retrieval. 2 Recall: Term-document matrix This matrix is the basis for computing the similarity between documents and queries. Today:
Text Categorization Moshe Koppel Lecture 12:Latent Semantic Indexing Adapted from slides by Prabhaker Raghavan, Chris Manning and TK Prasad.
SINGULAR VALUE DECOMPOSITION (SVD)
Orthogonalization via Deflation By Achiya Dax Hydrological Service Jerusalem, Israel
Dimension reduction : PCA and Clustering Slides by Agnieszka Juncker and Chris Workman modified by Hanne Jarmer.
Direct Methods for Sparse Linear Systems Lecture 4 Alessandra Nardi Thanks to Prof. Jacob White, Suvranu De, Deepak Ramaswamy, Michal Rewienski, and Karen.
V. Clustering 인공지능 연구실 이승희 Text: Text mining Page:82-93.
Matrix Factorization and its applications By Zachary 16 th Nov, 2010.
Algebraic Techniques for Analysis of Large Discrete-Valued Datasets 
Algebraic Techniques for Analysis of Large Discrete-Valued Datasets  Mehmet Koyutürk and Ananth Grama, Dept. of Computer Sciences, Purdue University {koyuturk,
Machine Learning Supervised Learning Classification and Regression K-Nearest Neighbor Classification Fisher’s Criteria & Linear Discriminant Analysis Perceptron:
Auburn University
Dimensionality Reduction and Principle Components Analysis
PREDICT 422: Practical Machine Learning
Compressive Coded Aperture Video Reconstruction
Auburn University COMP7330/7336 Advanced Parallel and Distributed Computing Mapping Techniques Dr. Xiao Qin Auburn University.
Lecture 19 MA471 Fall 2003.
Applying Twister to Scientific Applications
LSI, SVD and Data Management
Jianping Fan Dept of CS UNC-Charlotte
Singular Value Decomposition
Chapter 6. Large Scale Optimization
Algebraic Techniques for Analysis of Large Discrete-Valued Datasets
CSCE569 Parallel Computing
The BIRCH Algorithm Davitkov Miroslav, 2011/3116
Parallelization of Sparse Coding & Dictionary Learning
Concept Decomposition for Large Sparse Text Data Using Clustering
Outline Singular Value Decomposition Example of PCA: Eigenfaces.
Birch presented by : Bahare hajihashemi Atefeh Rahimi
Lecture 13: Singular Value Decomposition (SVD)
To accompany the text “Introduction to Parallel Computing”,
Restructuring Sparse High Dimensional Data for Effective Retrieval
Lecture 20 SVD and Its Applications
Chapter 6. Large Scale Optimization
Presentation transcript:

Parallelism in High-Performance Computing Applications Exploit parallelism through the entire simulation/computation pipeline from I/O to visualization. Current approaches have taken isolated approaches to parallel applications, data archival, retrieval, analysis, and visualization. In addition to our work on parallel computing, we have also investigated topics in parallel/distributed visualization, data analysis, and compression.

Scalable Parallel Volume Visualization Highly optimized shear-warp algorithm forms the basis for parallelization. Optimizations include image and object space coherence, early termination, compression. Parallel (MPI-based) formulation on SP is shown to scale to 128 processors and achieve frame rates in excess of 15 fps for UNC Brain dataset (256x256x167).

Parallel Shear-Warp Data Partitioning: Compositing: Load Balancing: Sheared volume partitioning Compositing: Software compositing/binary aggregation Load Balancing: Coherence in object movement -- use previous frame to load balance current frame.

Data/Computation Partitioning

Performance Notes Only scan-lines corresponding to incremental shear need to be communicated between frames. Since relative shear is not large, this communication overhead is small.

Performance Notes MPI version tested on up to 128 processors of an IBM SP (112MHz PowerPC 604), among other platforms. Datasets scaling from 128 x 128 x 84 to 256 x 256 x 167 (UNC Brain/Head datasets).

Performance Notes. All rendering times are in milliseconds and include compositing time.

Data Analysis Techniques for Very High Dimensional Data Datasets from simulations/physical processes can have extremely high dimensionality and large volume. This data is also typically sparse. Interpreting this data requires scalable techniques for detection of dominant and deviant patterns. Handling large discrete-valued datasets Extracting co-occurrences between events Summarizing data in an error-bounded fashion Finding concise representations for summary data

Background Singular Value Decomposition (SVD) [Berry et.al., 1995] Decompose matrix into A=USVT U and V orthogonal matrices, S diagonal with singular values Used for Latent Semantic Indexing in Information Retrieval Truncate decomposition to compress data

Background Semi-Discrete Decomposition (SDD) [Kolda and O’Leary, 1998] Restrict entries of U and V to {-1,0,1} Requires very small amount of storage Can perform as well as SVD in LSI using less than one-tenth the storage Effective in finding outlier clusters works well for datasets containing a large number of small clusters

Rank-1 Approximations x : presence vector y : pattern vector

Discrete Rank-1 Approximation Problem: Given discrete matrix Amxn , find discrete vectors xmx1 and ynx1 to Minimize = number of non-zeros in the error matrix Heuristic: Fix y, set solve for x to Maximize Iteratively solve for x and y until no improvement possible

Recursive Algorithm - At any step, given rank-one approximation AxyT, split A into A1 and A0 based on rows : - if x(i)=0 row i goes to A0 - if x(i)=1 row i goes to A1 - Stop when - Hamming radius of A1, maximum of the Hamming distances of A1pattern vector, is less then some threshold - All rows of A are present in A1 (if A1does not satisfy Hamming radius condition, can split A1 based on Hamming distances)

Effectiveness of Analysis

Effectiveness of Analysis

Run-time Scalability Rank-1 approximation requires O(nz(A)) time Total run-time at each level in the recursive tree cannot exceed this since total # of nonzeros at each level is at most nz(A)  Run-time is linear in nz(A) runtime vs # columns runtime vs # rows runtime vs # nonzeros