A Practical Randomized CP Tensor Decomposition

Size: px
Start display at page:

Download "A Practical Randomized CP Tensor Decomposition"

Transcription

1 A Practical Randomized CP Tensor Decomposition Casey Battaglino, Grey Ballard 2, and Tamara G. Kolda 3 SIAM CSE 27 Atlanta, GA Georgia Tech Computational Sci. and Engr. Wake Forest University Sandia National Labs cbattaglino3@gatech.edu 2 ballard@wfu.edu 3 tgkolda@sandia.gov

2 Tensors modes I3 I I4 I4 etc I2 I5 order order 2 order 3 order 4 order 5 order N 2

3 Problem Matrix Decomposition ` ` `... (SVD) 3

4 Problem Matrix Decomposition ` ` `... (SVD) Tensor Decomposition ` ` `... (CP) 4

5 CP: Multi-Way Data Analysis PCA (2D) A neurons data matrix principal components analysis (PCA) = reconstruction trial phase B trial # trial #5 trial # neuron factors phase factors trial factors CP (3D) functional cell types trial start Canonical Polyadic (CP) tensor decomposition within trial dynamics trial end first trial learning dynamics reconstruction last trial neurons phase trials + + = (source: Alex Williams, Stanford/Sandia) 5

6 Standard Method Alternating Least Squares (CP-ALS) ` ` ` ` ` ` mode mode 2 mode 3 Solve a sequence of least squares problems, one for each mode, repeat A () A (2) A (3) 6

7 Our Method Randomized Least Squares or sketching 7

8 Contributions We show how randomized sketching techniques can extend from matrices to general dense tensors We demonstrate a novel randomized least squares algorithm for the CP decomposition, with a novel stopping condition This enables us to scale to much larger data sets In addition to speed, there is evidence that this randomization increases robustness! 8

9 Example: Linear Regression b A ` " y 6 px 3,y 3 q m y 3 x 3 2 n `" n x Find line 2x ` satisfying: min }A b} 9

10 Example: Linear Regression b A ` " y 6 m y 3 x 3 2 n `" y 2 ` x 2 n Method of Normal Equations Solves for : x min }A b} pa T Aq A T A pa T Aq A T b A : b

11 Example: Linear Regression b A ` " y 6 m y 3 x 3 2 n `" y 2 ` x 2 n In MATLAB: x min }A b} Azb

12 Sampling Linear Regression What if we sample points? (Uniform, Random) 6 row-sampling operator 5 4 Sb SA 3 2 s 2 n

13 Sampling Linear Regression Sample A and b (MATLAB): >> srows = randi(m,s,); %sample s rows w. replacement >> As = A(srows, :); >> bs = b(srows); >> x = A \ b; >> xs = As \ bs; 6 row-sampling operator 5 4 y 2 ` x s Sb SA n

14 Sampling Linear Regression Chol: Opmn 2 ` n 3 q flops QR: Opmn 2 q flops b A 6 m " n m y 3 x y 2 ` x Chol: Opsn 2 ` n 3 q flops QR: Opsn 2 q flops m " s n s Sb SA (Fast Solution, Approximate) y ˆ2 ` ˆx n

15 Danger: Undetermined Sampling A SA SA may be rank-deficient > underdetermined! 5

16 Underdetermined Sampling m A x b n c n m Given full-rank A: Consider a row containing the only nonzero in its column 6

17 Underdetermined Sampling m A x b n c n m s SA n ˆx ˆb Any sampling must contain that row or be rank-deficient. s Given full-rank A: Consider a row containing the only nonzero in its column Uniform sampling w/ replacement: s = O(mlogm) samples needed (so s > m; sampling is useless here!) (Avron, Maymounkov & Toledo 2) 7

18 Mixing A x b c n?? n Idea: Transform A,b before sampling m (Ailon & Chazelle, 26 / Drineas, Mahoney, Muthukrishnan, Sarlós 27 / Rokhlin & Tygert 28) 8

19 Mixing Trick: mix up the rows (good)

20 Mixing Trick: mix up the rows F problem with vectors that are sparse in the frequency domain (bad) 8 2

21 random sign flipping helps spread out smooth vectors D Mixing Trick: mix up the rows F (good)

22 Mixing Trick: mix up the rows Fast Johnson-Lindenstrauss Transform (Ailon & Chazelle, 26) Randomly sign-flip rows, then perform one of: FFT WHT DHT F F D A x F - - D - - b Orthogonal transformations; Change of basis 22

23 Mixing Trick: mix up the rows Fast Johnson-Lindenstrauss Transform (Ailon & Chazelle, 26) Randomly sign-flip rows, then perform one of: FFT WHT DHT F F D A x F - - D - - b Opmn log mq  Opmnq Opm log mq ˆb Opmq 23

24 Mixing Fast Johnson-Lindenstrauss Transform (Ailon & Chazelle, 26) Randomly sign-flip rows, then perform one of: Trick: mix up the rows - - FFT WHT DHT (other options include DCT, random Givens rotations, random orthogonal matrix) F D A x F F D b Opmn log mq  Opmnq Opm log mq ˆb Opmq 24

25 Mixing Fast Johnson-Lindenstrauss Transform (Ailon & Chazelle, 26) Randomly sign-flip rows, then perform one of: Trick: mix up the rows S Â x FFT WHT DHT (other options include DCT, random Givens rotations, random orthogonal matrix) S ˆb F We should then be able to sample a small number of rows 25

26 Coherence Given a matrix Q whose columns are an orthonormal basis for A: e.g. rq, Rs QRpAq Coherence is the maximum leverage score: µpaq max i }Q i, } 2 2 (measuring correlation with the standard basis) e.g. Well-mixed matrix n m µpaq e.g. Identity matrix incoherent coherent 26

27 n #Samples (to recover a full-rank preconditioner) n m µpaq incoherent coherent m µpaq n m Ñ s Opn log nq (tiny) µpaq Ñ s Opm log mq (larger than input) (Avron, Maymounkov & Toledo 2) 27

28 Faster Approximate Least Squares mix sample min x }Ax b} 2 min x }FDAx FDb} 2 use as approximate solution for original problem solve exactly min x }SFDAx SFDb} 2 ˆx SFDAzSFDb (Drineas, Mahoney, Muthukrishnan, Sarlós 27/) 28

29 Sketching min x }Ax b} 2 min x }SFDAx SFDb} 2 Has found practical uses, e.g. Blendenpik : Precondition using Randomized Least Squares Then apply a standard iterative method (LSQR) ~4X overall speedup over LAPACK (Avron, Maymounkov & Toledo 2) 29

30 Sketching Can we apply this technique to tensor applications? I3 I I2 3

31 (our algorithms): CPRAND CPRAND-FFT Yes Fit CP-ALS(R) (standard algorithms) CP-ALS(H) x3x3 Tensor, R=5, Noise = % Time(s) 3

32 CP Decomposition X «X Rÿ r a r b r c r SVD: ` ` `... 3D CP: ` ` `... R number of components 32

33 Deriving CP-ALS I K X «X Rÿ r c b a r b r c r ` ` `... J a R Store factors in matrix form: R R I A J B K R C a b c 33

34 K Tensor Fibers K K I I I J mode- fibers J mode-2 fibers J mode-3 fibers 34

35 K Matricization K aka unfolding K I I I J mode- fibers J mode-2 fibers J mode-3 fibers JK IK IJ I J K X pq X p2q X p3q 35

36 Kronecker Product A b B» fi a B a n B..... fl, a m B a mn B B, B,2 B, B,2 A, A,2 b B, B,2 A, A,2 B2, B2,2 B2, B2,2 A2, A2,2 B2, B2,2 B, B,2 B, B,2 A2, A2,2 B2, B2,2 B2, B2,2 every pair of scalar entries multiplied, in a block structure 36

37 CP Decomposition & Khatri-Rao Product The CP Representation: Has matricized form: X «X Rÿ r a r b r c r X pq «ApC d Bq T A d B ra b b, a 2 b b 2,...,a K b b K s c b c b b C B C d B b d columnwise kronecker 37

38 CP Decomposition c b a ` ` (order 3, extends to order N) R R I A J B K R C R Matricized form: X pq «ApC d Bq T Allows us to express the least squares problem: min A }X pq ApC d Bq T } F unknown coefficient matrix 38

39 Alternating Least Squares min A }X pq ApC d Bq T } F unknown coefficient matrix Solve for each factor A, B, C in alternating fashion: A X pq { pc d Bq T (Transpose least squares, so forward-slash) 39

40 Alternating Least Squares order 3: A X p2q { pc d Bq T X pq repeat B X p2q { pc d Aq T C X p3q { pb d Aq T R ˆ R A X p2q pc d Bq { pc T C B T Bq X pq B X p2q pc d Aq { pc T C A T Aq C X p3q pb d Aq { pb T B A T Aq special structure 4

41 CP-ALS Algorithm CP-ALS : procedure CP-ALS(X,R) ô X P R IˆJˆK 2: Initialize factor matrices B, C 3: repeat 4: A X pq rpc d Bq T s : X pq pc d BqpC T C B T Bq : 5: B X p2q rpc d Aq T s : X p2q pc d AqpC T C A T Aq : 6: C X p3q rpb d Aq T s : X p3q pb d AqpB T B A T Aq : 7: until termination criteria met 8: return factor matrices A, B, C 9: end procedure init: random or nvec terminate when fit stops changing: }X X} }X} 4

42 CPRAND A X pq { pc d Bq T A X pq S T { SpC d Bq T JK JK I { R X pq pc d Bq T CP-RAND: Just sample X pq and pc d Bq T 42

43 CPRAND A X pq { pc d Bq T A X pq S T { SpC d Bq T JK JK I { R X pq pc d Bq T CP-RAND: Just sample X pq and pc d Bq T... and hope that pc d Bq T is incoherent... 43

44 Algorithm CPRAND CPRAND : procedure CPRAND(X,R,S) ô X P R IˆJˆK 2: Initialize factor matrices B, C 3: repeat 4: Define uniform random sampling operators S A, S B, S C 5: A X pq S T A ps ApC d Bq T q : 6: B X p2q S T B ps BpC d Aq T q : 7: C X p3q S T C ps CpB d Cq T q : 8: until termination criteria met 9: return A, B, C : end procedure (Order 3 Example, extends to arbitrary order) 44

45 Algorithm CPRAND CPRAND : procedure CPRAND(X,R,S) ô X P R IˆJˆK 2: Initialize factor matrices B, C 3: repeat 4: Define uniform random sampling operators S A, S B, S C 5: A X pq S T A ps ApC d Bq T q : 6: B X p2q S T B ps BpC d Aq T q : 7: C X p3q S T C ps CpB d Cq T q : 8: until termination criteria met 9: return A, B, C : end procedure (Order 3 Example, extends to arbitrary order) Trick : Instead of matricizing X and then sampling, sample fibers from X and then matricize the result. 45

46 Algorithm CPRAND CPRAND : procedure CPRAND(X,R,S) ô X P R IˆJˆK 2: Initialize factor matrices B, C 3: repeat 4: Define uniform random sampling operators S A, S B, S C 5: A X pq S T A pskrps A, C, Bq T q : X pq S T A ps ApC d Bq T q : 6: B X p2q S T B pskrps B, C, Aq T q : X p2q S T B ps BpC d Aq T q : 7: C X p3q S T C pskrps C, B, Aq T q : X p3q S T C ps CpB d Cq T q : 8: until termination criteria met 9: return A, B, C : end procedure (Order 3 Example, extends to arbitrary order) SKR: Sample KRP without forming Trick 2: We don t need to form full KR-Product j i I B i R C j R B c i b * j S I C S(C B) 46

47 Algorithm CPRAND CPRAND : procedure CPRAND(X,R,S) ô X P R IˆJˆK 2: Initialize factor matrices B, C 3: repeat 4: Define uniform random sampling operators S A, S B, S C 5: A X pq S T A pskrps A, C, Bq T q : X pq S T A ps ApC d Bq T q : 6: B X p2q S T B pskrps B, C, Aq T q : X p2q S T B ps BpC d Aq T q : 7: C X p3q S T C pskrps C, B, Aq T q : X p3q S T C ps CpB d Cq T q : 8: until termination criteria met 9: return A, B, C : end procedure (Order 3 Example, extends to arbitrary order) No mixing performed here, yet converges in many cases. Our result: µpc d Bq µpcqµpbq 47

48 Coherence of Kronecker Product identity: AB b CD pabcqpb b Dq Lemma : µpa b Bq µpaqµpbq Proof: A b B Q A R A b Q B R B pq A b Q B qpr A b R B q Q AbB R AbB Each row of Q AbB has form Q A pi, :qbq B pj, :q Proof follows from }a b b} }a}}b} 48

49 Coherence of KR Product identity: AB d CD pa b CqpB d Dq Lemma 2: µpa d Bq µpaqµpbq Proof: pq A b Q B qpr A d R B q pq A b Q B qq R R R Q AdB R AdB. Q AdB R AdB ˆq T i...row i of Q A b Q B ˆli ˆq T i Q R Q T Rˆq i Q T R ˆq i 49

50 Mixing Tensors For better convergence & guarantees, we need to mix Goals: Space Efficient (A single mixed tensor, avoid forming full KRP) Time Efficient (Minimize mixing costs, sample sizes) 5

51 Mixing Tensors CPRAND without mixing was: A X pq S T { SpC d Bq T Naive approach: mix X pq, C d B directly, then sample. I JK X pq JK R C d B (Expensive) 5

52 Mixing Factors Can we avoid forming the full KRP? R R I A J B K C JK R R C d B 52

53 Mixing Factors CPRAND without mixing was: A X pq S T { SpC d Bq T R R R R I A J B K C s Trick 3: Transform individual factors pf C D C C d F B D B Bq Sample Khatri Rao product without fully forming 53

54 Mixing Factors Fact: AB d CD pa b CqpB d Dq So the mixed factors expand: pf C D C C d F B D B Bq pf C D C b F B D B qpc d Bq apply to X pq (the LHS) X pq pf C D C b F B D B q 54

55 Mode-n Multiplication I K X ˆn U UX pnq K K IJ ˆ3 U U ˆ K J mode-3 fibers I A X pq pf C D C b F B D B q X ˆ F C D C ˆ2 F B D B ˆ3 I A 55

56 Mixing Factors Trick 4: Mix original tensor only once: ˆX X ˆ3 F C D C ˆ2 F B D B ˆ F A D A matricize in one mode ˆX pq D A F A X pq pf C D C b F B D B q sample rows D A F A X pq pf C D C b F B D B qs T solve least squares  D A F A X pq pf C D C b F B D B qs T psz pqt q : inverse mix the final, small, solution A D A F A... 56

57 Mixing Factors Trick 4: Mix original tensor only once: ˆX X ˆ3 F C D C ˆ2 F B D B ˆ F A D A matricize in one mode ˆX pq D A F A X pq pf C D C b F B D B q sample rows D A F A X pq pf C D C b F B D B qs T solve least squares  D A F A X pq pf C D C b F B D B qs T psz pqt q : inverse mix the final, small, solution A D A F A... 57

58 Mixing Factors Trick 4: Mix original tensor only once: ˆX X ˆ3 F C D C ˆ2 F B D B ˆ F A D A matricize in one mode ˆX pq D A F A X pq pf C D C b F B D B q sample rows D A F A X pq pf C D C b F B D B qs T solve least squares  D A F A X pq pf C D C b F B D B qs T psz pqt q : inverse mix the final, small, solution A D A F A... 58

59 Mixing Factors Trick 4: Mix original tensor only once: ˆX X ˆ3 F C D C ˆ2 F B D B ˆ F A D A matricize in one mode ˆX pq D A F A X pq pf C D C b F B D B q sample rows D A F A X pq pf C D C b F B D B qs T solve least squares  D A F A X pq pf C D C b F B D B qs T psz pqt q : inverse mix the final, small, solution A D A F A... 59

60 Mixing Factors Trick 4: Mix original tensor only once: ˆX X ˆ3 F C D C ˆ2 F B D B ˆ F A D A matricize in one mode ˆX pq D A F A X pq pf C D C b F B D B q sample rows D A F A X pq pf C D C b F B D B qs T solve least squares  D A F A X pq pf C D C b F B D B qs T psz pqt q : inverse mix the final, small, solution A D A F A  6

61 Mixing Factors Trick 4: Mix original tensor only once: ˆX X ˆ3 F C D C ˆ2 F B D B ˆ F A D A Summary: inverse mix small result at end solve small system of size S * R A pnq D n F n p ˆX pnq S T qrpsẑ pnq q s : efficiently sample S fibers from mode n of ˆX before reshaping (Trick ) efficiently sample S rows of KRP (Trick 2) mix individual factors (Trick 3) 6

62 Algorithm 3: CPRAND-MIX Algorithm CPRAND (Order 3 Example, extends to arbitrary order) : procedure CPRAND(X,R,S) ô X P R IˆJˆK 2: Mix: ˆX Xˆ3F C D C ˆ2 F B D B ˆ F A D A 3: Initialize factor matrices B, C 4: repeat 5: Define uniform random sampling operators S A, S B, S C ı 6: A D A F A ˆX pq S T A pskrps A, F C D C C, F B D B Bq T q : ı 7: B D B F B ˆX p2q S T B pskrps B, F C D C C, F A D A Aq T q : ı 8: C D C F C ˆX p3q S T C pskrps C, F B D B B, F C D C Cq T q : 9: until termination criteria met : return A, B, C : end procedure arg min kax bk 2 = arg min x2r n x2r n apple <(A) =(A) x apple <(b) =(b) 2 (Trick 5) 62

63 Algorithm 4: CPRAND-PREMIX If we use a real-valued orthogonal transform such as the DCT, We can simply mix, call CPRAND, then unmix the output. Algorithm CPRAND-PREMIX : function CPRAND-PREMIX(X,R,S). X 2 R I I N 2: Define random sign-flip operators D m and orthogonal matrices F m,m2{,...,n} 3: 4: Mix: ˆX X F D N F N D N [, {Â (n) }]= CPRAND(ˆX,R,S) 5: for n =,...,N do 6: Unmix: A (n) = D n Fn TÂ(n) 7: end for 8: return factor matrices { A (n) } 9: end function 63

64 Trick 5: Stopping Criterion I K J CP-ALS: }X X} }X} expensive inner product b }X} 2 `} X} 2 2 X, X }X} 64

65 Trick 5: Stopping Criterion I K J CP-ALS: }X X} }X} expensive inner product b }X} 2 `} X} 2 2 X, X }X} I CPRAND: K Uniformly sample : X I Ä I risbrjsbrks Choose some subset of tensor entries J 65

66 Trick 5: Stopping Criterion I K J CP-ALS: }X X} }X} expensive inner product b }X} 2 `} X} 2 2 X, X }X} I J CPRAND: K Uniformly sample : X }X X} }X} I Ä I risbrjsbrks a I mean t e 2 i i P I u }X} «approximate fit a I mean t e 2 i i P I u }X} 66

67 Trick 5: Stopping Criterion I J CPRAND: K Uniformly sample : X }X X} }X} I Ä I risbrjsbrks a I mean t e 2 i i P I u }X} «approximate fit a I mean t e 2 i i P I u }X} How many samples are needed? Since we are approximating a mean, we can apply the Chernoff-Hoeffding bound 67

68 Trick 5: Stopping Criterion K approximate fit I }X X} }X} a I mean t e 2 i i P I u }X} «a I mean t e 2 i i P I u }X} J Very conservative bound: (assuming e i is i.i.d) (µ max = max(e 2 i )) #samples ˆP 372µ 2 max Yields 95% confidence that the sampled fit is within 5% of true fit In practice: ˆP =2 4 yields error < 3 on synthetic data 68

69 Trick 5: Stopping Criterion K approximate fit I }X X} }X} a I mean t e 2 i i P I u }X} «a I mean t e 2 i i P I u }X} J Terminate after fit ceases to decrease after a certain number of iterations 69

70 Experimental Setup Environment MATLAB R26a & Tensor Toolbox v2.6. Intel Xeon E5-265 Ivy Bridge 2. GHz machine with 32 GB of memory Data Synthetic random tensor Factors of rank Rtrue Factors are generated with collinearity Observation is combined with Gaussian noise Score is a measure of how well the original factors are recovered, from to 7

71 Example Run True Rank = 5, Target Rank = 5, Noise = %, Collinearity.9 7

72 Per-Iteration Times R 5 S R log R 72

73 Stopping Criterion Speedup (stopping condition with ˆP =2 4 ) 73

74 Preprocessing Times nvec on each mode-n unfolding vs. fft on fibers of each mode 74

75 Experiment : order Time (s) Score Fit Iterations until Termination CPRAND MIX CPRAND CP ALS(R) CP ALS(H) Size: 9x9x9x9 Rtrue: 5 R: {5,6} Noise: % 6 Runs Collinearity: {.5,.9} 75

76 Experiment : order Time (s) Score Fit Iterations until Termination CPRAND MIX CPRAND CP ALS(R) CP ALS(H) Size: 9x9x9x9 Rtrue: 5 R: {5,6} Noise: % 6 Runs Collinearity: {.5,.9} 76

77 Conclusion (Experiment ) CPRAND can surprisingly outperform CP-ALS in both time and factor recovery (given the same termination criteria). CPRAND has very low per-iteration time, but may take more iterations until termination. CPRAND-FFT takes fewer iterations and samples at the cost of pre-processing time. 77

78 Experiment 2: Hazardous Gas Classification We replicate an experiment from Vervliet & Lauthauwer, 26: 72 sensors 25,9 time steps 3 experiments per gas (3 gases: CO, Acetaldehyde, Ammonia) 259 x 72 x 9 Tensor 3.4 GB, dense sensor time experiment Goal: Classify which gas is released in each experiment Here we use CPRAND without mixing! 78

79 Experiment 2: Hazardous Gas Classification (over trials) Method Median Time (s) Median Fit Median Classification Error CPRAND % CP-ALS(H) % CP-ALS(R) % S =, ˆP =2 4 Factors Recovered by CPRAND: 79

80 Experiment 3: COIL- Image Set We replicate an experiment from Zhou, Cichocki, and Xie (23): COIL-: objects, images of each from 72 different poses. Each image is RGB with size 28x28 pixels. 28 x 28 x 3 x 72 Tensor (2.8 GB) image size color channel image id 8

81 Experiment 3: COIL- Image Set Here we use CPRAND-FFT (mixing): # Samples (S) Median Speedup Median Fit trials, R = 2, ˆP =2 4 8

82 Future Work Randomization in the Tucker model Randomization in sparse CP (for scalability) 82

83 Contact Casey Battaglino 83

A Practical Randomized CP Tensor Decomposition

A Practical Randomized CP Tensor Decomposition A Practical Randomized CP Tensor Decomposition Casey Battaglino, Grey Ballard 2, and Tamara G. Kolda 3 SIAM AN 207, Pittsburgh, PA Georgia Tech Computational Sci. and Engr. 2 Wake Forest University 3 Sandia

More information

A fast randomized algorithm for overdetermined linear least-squares regression

A fast randomized algorithm for overdetermined linear least-squares regression A fast randomized algorithm for overdetermined linear least-squares regression Vladimir Rokhlin and Mark Tygert Technical Report YALEU/DCS/TR-1403 April 28, 2008 Abstract We introduce a randomized algorithm

More information

A fast randomized algorithm for approximating an SVD of a matrix

A fast randomized algorithm for approximating an SVD of a matrix A fast randomized algorithm for approximating an SVD of a matrix Joint work with Franco Woolfe, Edo Liberty, and Vladimir Rokhlin Mark Tygert Program in Applied Mathematics Yale University Place July 17,

More information

Efficient CP-ALS and Reconstruction From CP

Efficient CP-ALS and Reconstruction From CP Efficient CP-ALS and Reconstruction From CP Jed A. Duersch & Tamara G. Kolda Sandia National Laboratories Livermore, CA Sandia National Laboratories is a multimission laboratory managed and operated by

More information

Sketching as a Tool for Numerical Linear Algebra

Sketching as a Tool for Numerical Linear Algebra Sketching as a Tool for Numerical Linear Algebra David P. Woodruff presented by Sepehr Assadi o(n) Big Data Reading Group University of Pennsylvania February, 2015 Sepehr Assadi (Penn) Sketching for Numerical

More information

Randomized Algorithms for Matrix Computations

Randomized Algorithms for Matrix Computations Randomized Algorithms for Matrix Computations Ilse Ipsen Students: John Holodnak, Kevin Penner, Thomas Wentworth Research supported in part by NSF CISE CCF, DARPA XData Randomized Algorithms Solve a deterministic

More information

A fast randomized algorithm for orthogonal projection

A fast randomized algorithm for orthogonal projection A fast randomized algorithm for orthogonal projection Vladimir Rokhlin and Mark Tygert arxiv:0912.1135v2 [cs.na] 10 Dec 2009 December 10, 2009 Abstract We describe an algorithm that, given any full-rank

More information

A fast randomized algorithm for the approximation of matrices preliminary report

A fast randomized algorithm for the approximation of matrices preliminary report DRAFT A fast randomized algorithm for the approximation of matrices preliminary report Yale Department of Computer Science Technical Report #1380 Franco Woolfe, Edo Liberty, Vladimir Rokhlin, and Mark

More information

Blendenpik: Supercharging LAPACK's Least-Squares Solver

Blendenpik: Supercharging LAPACK's Least-Squares Solver Blendenpik: Supercharging LAPACK's Least-Squares Solver The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters. Citation As Published Publisher

More information

Compressive Sensing of Sparse Tensor

Compressive Sensing of Sparse Tensor Shmuel Friedland Univ. Illinois at Chicago Matheon Workshop CSA2013 December 12, 2013 joint work with Q. Li and D. Schonfeld, UIC Abstract Conventional Compressive sensing (CS) theory relies on data representation

More information

A Fast Algorithm For Computing The A-optimal Sampling Distributions In A Big Data Linear Regression

A Fast Algorithm For Computing The A-optimal Sampling Distributions In A Big Data Linear Regression A Fast Algorithm For Computing The A-optimal Sampling Distributions In A Big Data Linear Regression Hanxiang Peng and Fei Tan Indiana University Purdue University Indianapolis Department of Mathematical

More information

Sketching as a Tool for Numerical Linear Algebra

Sketching as a Tool for Numerical Linear Algebra Sketching as a Tool for Numerical Linear Algebra (Part 2) David P. Woodruff presented by Sepehr Assadi o(n) Big Data Reading Group University of Pennsylvania February, 2015 Sepehr Assadi (Penn) Sketching

More information

Kronecker Product Approximation with Multiple Factor Matrices via the Tensor Product Algorithm

Kronecker Product Approximation with Multiple Factor Matrices via the Tensor Product Algorithm Kronecker Product Approximation with Multiple actor Matrices via the Tensor Product Algorithm King Keung Wu, Yeung Yam, Helen Meng and Mehran Mesbahi Department of Mechanical and Automation Engineering,

More information

High Performance Parallel Tucker Decomposition of Sparse Tensors

High Performance Parallel Tucker Decomposition of Sparse Tensors High Performance Parallel Tucker Decomposition of Sparse Tensors Oguz Kaya INRIA and LIP, ENS Lyon, France SIAM PP 16, April 14, 2016, Paris, France Joint work with: Bora Uçar, CNRS and LIP, ENS Lyon,

More information

A randomized block sampling approach to the canonical polyadic decomposition of large-scale tensors

A randomized block sampling approach to the canonical polyadic decomposition of large-scale tensors A randomized block sampling approach to the canonical polyadic decomposition of large-scale tensors Nico Vervliet Joint work with Lieven De Lathauwer SIAM AN17, July 13, 2017 2 Classification of hazardous

More information

Computational Linear Algebra

Computational Linear Algebra Computational Linear Algebra PD Dr. rer. nat. habil. Ralf-Peter Mundani Computation in Engineering / BGU Scientific Computing in Computer Science / INF Winter Term 2018/19 Part 6: Some Other Stuff PD Dr.

More information

Approximate Principal Components Analysis of Large Data Sets

Approximate Principal Components Analysis of Large Data Sets Approximate Principal Components Analysis of Large Data Sets Daniel J. McDonald Department of Statistics Indiana University mypage.iu.edu/ dajmcdon April 27, 2016 Approximation-Regularization for Analysis

More information

Sparse Solutions of Systems of Equations and Sparse Modelling of Signals and Images

Sparse Solutions of Systems of Equations and Sparse Modelling of Signals and Images Sparse Solutions of Systems of Equations and Sparse Modelling of Signals and Images Alfredo Nava-Tudela ant@umd.edu John J. Benedetto Department of Mathematics jjb@umd.edu Abstract In this project we are

More information

Tensor Decompositions and Applications

Tensor Decompositions and Applications Tamara G. Kolda and Brett W. Bader Part I September 22, 2015 What is tensor? A N-th order tensor is an element of the tensor product of N vector spaces, each of which has its own coordinate system. a =

More information

Randomized algorithms for the approximation of matrices

Randomized algorithms for the approximation of matrices Randomized algorithms for the approximation of matrices Luis Rademacher The Ohio State University Computer Science and Engineering (joint work with Amit Deshpande, Santosh Vempala, Grant Wang) Two topics

More information

CS 229r: Algorithms for Big Data Fall Lecture 17 10/28

CS 229r: Algorithms for Big Data Fall Lecture 17 10/28 CS 229r: Algorithms for Big Data Fall 2015 Prof. Jelani Nelson Lecture 17 10/28 Scribe: Morris Yau 1 Overview In the last lecture we defined subspace embeddings a subspace embedding is a linear transformation

More information

BLENDENPIK: SUPERCHARGING LAPACK'S LEAST-SQUARES SOLVER

BLENDENPIK: SUPERCHARGING LAPACK'S LEAST-SQUARES SOLVER BLENDENPIK: SUPERCHARGING LAPACK'S LEAST-SQUARES SOLVER HAIM AVRON, PETAR MAYMOUNKOV, AND SIVAN TOLEDO Abstract. Several innovative random-sampling and random-mixing techniques for solving problems in

More information

Introduction to Compressed Sensing

Introduction to Compressed Sensing Introduction to Compressed Sensing Alejandro Parada, Gonzalo Arce University of Delaware August 25, 2016 Motivation: Classical Sampling 1 Motivation: Classical Sampling Issues Some applications Radar Spectral

More information

Linear Algebra Review. Vectors

Linear Algebra Review. Vectors Linear Algebra Review 9/4/7 Linear Algebra Review By Tim K. Marks UCSD Borrows heavily from: Jana Kosecka http://cs.gmu.edu/~kosecka/cs682.html Virginia de Sa (UCSD) Cogsci 8F Linear Algebra review Vectors

More information

Faloutsos, Tong ICDE, 2009

Faloutsos, Tong ICDE, 2009 Large Graph Mining: Patterns, Tools and Case Studies Christos Faloutsos Hanghang Tong CMU Copyright: Faloutsos, Tong (29) 2-1 Outline Part 1: Patterns Part 2: Matrix and Tensor Tools Part 3: Proximity

More information

Approximate Spectral Clustering via Randomized Sketching

Approximate Spectral Clustering via Randomized Sketching Approximate Spectral Clustering via Randomized Sketching Christos Boutsidis Yahoo! Labs, New York Joint work with Alex Gittens (Ebay), Anju Kambadur (IBM) The big picture: sketch and solve Tradeoff: Speed

More information

Methods for sparse analysis of high-dimensional data, II

Methods for sparse analysis of high-dimensional data, II Methods for sparse analysis of high-dimensional data, II Rachel Ward May 23, 2011 High dimensional data with low-dimensional structure 300 by 300 pixel images = 90, 000 dimensions 2 / 47 High dimensional

More information

Accelerated Dense Random Projections

Accelerated Dense Random Projections 1 Advisor: Steven Zucker 1 Yale University, Department of Computer Science. Dimensionality reduction (1 ε) xi x j 2 Ψ(xi ) Ψ(x j ) 2 (1 + ε) xi x j 2 ( n 2) distances are ε preserved Target dimension k

More information

Applications of Randomized Methods for Decomposing and Simulating from Large Covariance Matrices

Applications of Randomized Methods for Decomposing and Simulating from Large Covariance Matrices Applications of Randomized Methods for Decomposing and Simulating from Large Covariance Matrices Vahid Dehdari and Clayton V. Deutsch Geostatistical modeling involves many variables and many locations.

More information

PCA & ICA. CE-717: Machine Learning Sharif University of Technology Spring Soleymani

PCA & ICA. CE-717: Machine Learning Sharif University of Technology Spring Soleymani PCA & ICA CE-717: Machine Learning Sharif University of Technology Spring 2015 Soleymani Dimensionality Reduction: Feature Selection vs. Feature Extraction Feature selection Select a subset of a given

More information

Methods for sparse analysis of high-dimensional data, II

Methods for sparse analysis of high-dimensional data, II Methods for sparse analysis of high-dimensional data, II Rachel Ward May 26, 2011 High dimensional data with low-dimensional structure 300 by 300 pixel images = 90, 000 dimensions 2 / 55 High dimensional

More information

Institute for Computational Mathematics Hong Kong Baptist University

Institute for Computational Mathematics Hong Kong Baptist University Institute for Computational Mathematics Hong Kong Baptist University ICM Research Report 08-0 How to find a good submatrix S. A. Goreinov, I. V. Oseledets, D. V. Savostyanov, E. E. Tyrtyshnikov, N. L.

More information

14 Singular Value Decomposition

14 Singular Value Decomposition 14 Singular Value Decomposition For any high-dimensional data analysis, one s first thought should often be: can I use an SVD? The singular value decomposition is an invaluable analysis tool for dealing

More information

IV. Matrix Approximation using Least-Squares

IV. Matrix Approximation using Least-Squares IV. Matrix Approximation using Least-Squares The SVD and Matrix Approximation We begin with the following fundamental question. Let A be an M N matrix with rank R. What is the closest matrix to A that

More information

Parallel Tensor Compression for Large-Scale Scientific Data

Parallel Tensor Compression for Large-Scale Scientific Data Parallel Tensor Compression for Large-Scale Scientific Data Woody Austin, Grey Ballard, Tamara G. Kolda April 14, 2016 SIAM Conference on Parallel Processing for Scientific Computing MS 44/52: Parallel

More information

Randomized algorithms for the low-rank approximation of matrices

Randomized algorithms for the low-rank approximation of matrices Randomized algorithms for the low-rank approximation of matrices Yale Dept. of Computer Science Technical Report 1388 Edo Liberty, Franco Woolfe, Per-Gunnar Martinsson, Vladimir Rokhlin, and Mark Tygert

More information

Illustration by Chris Brigman Michigan Institute for Data Science (MIDAS) Seminar November 9, 2015

Illustration by Chris Brigman Michigan Institute for Data Science (MIDAS) Seminar November 9, 2015 Illustration by Chris Brigman Michigan Institute for Data Science (MIDAS) Seminar November 9, 2015 Sandia National Laboratories is a multi-mission laboratory managed and operated by Sandia Corporation,

More information

Lecture 12: Randomized Least-squares Approximation in Practice, Cont. 12 Randomized Least-squares Approximation in Practice, Cont.

Lecture 12: Randomized Least-squares Approximation in Practice, Cont. 12 Randomized Least-squares Approximation in Practice, Cont. Stat60/CS94: Randomized Algorithms for Matrices and Data Lecture 1-10/14/013 Lecture 1: Randomized Least-squares Approximation in Practice, Cont. Lecturer: Michael Mahoney Scribe: Michael Mahoney Warning:

More information

A Multi-Affine Model for Tensor Decomposition

A Multi-Affine Model for Tensor Decomposition Yiqing Yang UW Madison breakds@cs.wisc.edu A Multi-Affine Model for Tensor Decomposition Hongrui Jiang UW Madison hongrui@engr.wisc.edu Li Zhang UW Madison lizhang@cs.wisc.edu Chris J. Murphy UC Davis

More information

Linear Algebra (Review) Volker Tresp 2018

Linear Algebra (Review) Volker Tresp 2018 Linear Algebra (Review) Volker Tresp 2018 1 Vectors k, M, N are scalars A one-dimensional array c is a column vector. Thus in two dimensions, ( ) c1 c = c 2 c i is the i-th component of c c T = (c 1, c

More information

RandNLA: Randomization in Numerical Linear Algebra

RandNLA: Randomization in Numerical Linear Algebra RandNLA: Randomization in Numerical Linear Algebra Petros Drineas Department of Computer Science Rensselaer Polytechnic Institute To access my web page: drineas Why RandNLA? Randomization and sampling

More information

CSC 576: Variants of Sparse Learning

CSC 576: Variants of Sparse Learning CSC 576: Variants of Sparse Learning Ji Liu Department of Computer Science, University of Rochester October 27, 205 Introduction Our previous note basically suggests using l norm to enforce sparsity in

More information

Nonnegative Tensor Factorization using a proximal algorithm: application to 3D fluorescence spectroscopy

Nonnegative Tensor Factorization using a proximal algorithm: application to 3D fluorescence spectroscopy Nonnegative Tensor Factorization using a proximal algorithm: application to 3D fluorescence spectroscopy Caroline Chaux Joint work with X. Vu, N. Thirion-Moreau and S. Maire (LSIS, Toulon) Aix-Marseille

More information

Linear Algebra Methods for Data Mining

Linear Algebra Methods for Data Mining Linear Algebra Methods for Data Mining Saara Hyvönen, Saara.Hyvonen@cs.helsinki.fi Spring 2007 The Singular Value Decomposition (SVD) continued Linear Algebra Methods for Data Mining, Spring 2007, University

More information

/16/$ IEEE 1728

/16/$ IEEE 1728 Extension of the Semi-Algebraic Framework for Approximate CP Decompositions via Simultaneous Matrix Diagonalization to the Efficient Calculation of Coupled CP Decompositions Kristina Naskovska and Martin

More information

Sketched Ridge Regression:

Sketched Ridge Regression: Sketched Ridge Regression: Optimization and Statistical Perspectives Shusen Wang UC Berkeley Alex Gittens RPI Michael Mahoney UC Berkeley Overview Ridge Regression min w f w = 1 n Xw y + γ w Over-determined:

More information

The QR Factorization

The QR Factorization The QR Factorization How to Make Matrices Nicer Radu Trîmbiţaş Babeş-Bolyai University March 11, 2009 Radu Trîmbiţaş ( Babeş-Bolyai University) The QR Factorization March 11, 2009 1 / 25 Projectors A projector

More information

Illustration by Chris Brigman IPAM Workshop: Big Data Meets Computation February 2, 2017

Illustration by Chris Brigman IPAM Workshop: Big Data Meets Computation February 2, 2017 Illustration by Chris Brigman IPAM Workshop: Big Data Meets Computation February 2, 2017 Sandia National Laboratories is a multi-mission laboratory managed and operated by Sandia Corporation, a wholly

More information

Compressed Sensing and Robust Recovery of Low Rank Matrices

Compressed Sensing and Robust Recovery of Low Rank Matrices Compressed Sensing and Robust Recovery of Low Rank Matrices M. Fazel, E. Candès, B. Recht, P. Parrilo Electrical Engineering, University of Washington Applied and Computational Mathematics Dept., Caltech

More information

Preconditioning. Noisy, Ill-Conditioned Linear Systems

Preconditioning. Noisy, Ill-Conditioned Linear Systems Preconditioning Noisy, Ill-Conditioned Linear Systems James G. Nagy Emory University Atlanta, GA Outline 1. The Basic Problem 2. Regularization / Iterative Methods 3. Preconditioning 4. Example: Image

More information

Linear Systems of n equations for n unknowns

Linear Systems of n equations for n unknowns Linear Systems of n equations for n unknowns In many application problems we want to find n unknowns, and we have n linear equations Example: Find x,x,x such that the following three equations hold: x

More information

Numerical Methods. Elena loli Piccolomini. Civil Engeneering. piccolom. Metodi Numerici M p. 1/??

Numerical Methods. Elena loli Piccolomini. Civil Engeneering.  piccolom. Metodi Numerici M p. 1/?? Metodi Numerici M p. 1/?? Numerical Methods Elena loli Piccolomini Civil Engeneering http://www.dm.unibo.it/ piccolom elena.loli@unibo.it Metodi Numerici M p. 2/?? Least Squares Data Fitting Measurement

More information

Normalized power iterations for the computation of SVD

Normalized power iterations for the computation of SVD Normalized power iterations for the computation of SVD Per-Gunnar Martinsson Department of Applied Mathematics University of Colorado Boulder, Co. Per-gunnar.Martinsson@Colorado.edu Arthur Szlam Courant

More information

Randomly Sampling from Orthonormal Matrices: Coherence and Leverage Scores

Randomly Sampling from Orthonormal Matrices: Coherence and Leverage Scores Randomly Sampling from Orthonormal Matrices: Coherence and Leverage Scores Ilse Ipsen Joint work with Thomas Wentworth (thanks to Petros & Joel) North Carolina State University Raleigh, NC, USA Research

More information

Random Methods for Linear Algebra

Random Methods for Linear Algebra Gittens gittens@acm.caltech.edu Applied and Computational Mathematics California Institue of Technology October 2, 2009 Outline The Johnson-Lindenstrauss Transform 1 The Johnson-Lindenstrauss Transform

More information

Numerical Methods I Non-Square and Sparse Linear Systems

Numerical Methods I Non-Square and Sparse Linear Systems Numerical Methods I Non-Square and Sparse Linear Systems Aleksandar Donev Courant Institute, NYU 1 donev@courant.nyu.edu 1 MATH-GA 2011.003 / CSCI-GA 2945.003, Fall 2014 September 25th, 2014 A. Donev (Courant

More information

B553 Lecture 5: Matrix Algebra Review

B553 Lecture 5: Matrix Algebra Review B553 Lecture 5: Matrix Algebra Review Kris Hauser January 19, 2012 We have seen in prior lectures how vectors represent points in R n and gradients of functions. Matrices represent linear transformations

More information

Tighter Low-rank Approximation via Sampling the Leveraged Element

Tighter Low-rank Approximation via Sampling the Leveraged Element Tighter Low-rank Approximation via Sampling the Leveraged Element Srinadh Bhojanapalli The University of Texas at Austin bsrinadh@utexas.edu Prateek Jain Microsoft Research, India prajain@microsoft.com

More information

MAT 585: Johnson-Lindenstrauss, Group testing, and Compressed Sensing

MAT 585: Johnson-Lindenstrauss, Group testing, and Compressed Sensing MAT 585: Johnson-Lindenstrauss, Group testing, and Compressed Sensing Afonso S. Bandeira April 9, 2015 1 The Johnson-Lindenstrauss Lemma Suppose one has n points, X = {x 1,..., x n }, in R d with d very

More information

Preconditioning. Noisy, Ill-Conditioned Linear Systems

Preconditioning. Noisy, Ill-Conditioned Linear Systems Preconditioning Noisy, Ill-Conditioned Linear Systems James G. Nagy Emory University Atlanta, GA Outline 1. The Basic Problem 2. Regularization / Iterative Methods 3. Preconditioning 4. Example: Image

More information

CS168: The Modern Algorithmic Toolbox Lecture #8: How PCA Works

CS168: The Modern Algorithmic Toolbox Lecture #8: How PCA Works CS68: The Modern Algorithmic Toolbox Lecture #8: How PCA Works Tim Roughgarden & Gregory Valiant April 20, 206 Introduction Last lecture introduced the idea of principal components analysis (PCA). The

More information

1. Connecting to Matrix Computations. Charles F. Van Loan

1. Connecting to Matrix Computations. Charles F. Van Loan Four Talks on Tensor Computations 1. Connecting to Matrix Computations Charles F. Van Loan Cornell University SCAN Seminar October 27, 2014 Four Talks on Tensor Computations 1. Connecting to Matrix Computations

More information

Introduction to Tensors. 8 May 2014

Introduction to Tensors. 8 May 2014 Introduction to Tensors 8 May 2014 Introduction to Tensors What is a tensor? Basic Operations CP Decompositions and Tensor Rank Matricization and Computing the CP Dear Tullio,! I admire the elegance of

More information

Inverse of a Square Matrix. For an N N square matrix A, the inverse of A, 1

Inverse of a Square Matrix. For an N N square matrix A, the inverse of A, 1 Inverse of a Square Matrix For an N N square matrix A, the inverse of A, 1 A, exists if and only if A is of full rank, i.e., if and only if no column of A is a linear combination 1 of the others. A is

More information

Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images: Final Presentation

Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images: Final Presentation Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images: Final Presentation Alfredo Nava-Tudela John J. Benedetto, advisor 5/10/11 AMSC 663/664 1 Problem Let A be an n

More information

CS168: The Modern Algorithmic Toolbox Lecture #7: Understanding Principal Component Analysis (PCA)

CS168: The Modern Algorithmic Toolbox Lecture #7: Understanding Principal Component Analysis (PCA) CS68: The Modern Algorithmic Toolbox Lecture #7: Understanding Principal Component Analysis (PCA) Tim Roughgarden & Gregory Valiant April 0, 05 Introduction. Lecture Goal Principal components analysis

More information

Fast and Scalable Distributed Boolean Tensor Factorization

Fast and Scalable Distributed Boolean Tensor Factorization Fast and Scalable Distributed Boolean Tensor Factorization Namyong Park Seoul National University Email: namyong.park@snu.ac.kr Sejoon Oh Seoul National University Email: ohhenrie@snu.ac.kr U Kang Seoul

More information

Course Notes: Week 1

Course Notes: Week 1 Course Notes: Week 1 Math 270C: Applied Numerical Linear Algebra 1 Lecture 1: Introduction (3/28/11) We will focus on iterative methods for solving linear systems of equations (and some discussion of eigenvalues

More information

Statistical Pattern Recognition

Statistical Pattern Recognition Statistical Pattern Recognition Feature Extraction Hamid R. Rabiee Jafar Muhammadi, Alireza Ghasemi, Payam Siyari Spring 2014 http://ce.sharif.edu/courses/92-93/2/ce725-2/ Agenda Dimensionality Reduction

More information

arxiv: v3 [cs.ds] 21 Mar 2013

arxiv: v3 [cs.ds] 21 Mar 2013 Low-distortion Subspace Embeddings in Input-sparsity Time and Applications to Robust Linear Regression Xiangrui Meng Michael W. Mahoney arxiv:1210.3135v3 [cs.ds] 21 Mar 2013 Abstract Low-distortion subspace

More information

Matrix Factorization and Analysis

Matrix Factorization and Analysis Chapter 7 Matrix Factorization and Analysis Matrix factorizations are an important part of the practice and analysis of signal processing. They are at the heart of many signal-processing algorithms. Their

More information

The Canonical Tensor Decomposition and Its Applications to Social Network Analysis

The Canonical Tensor Decomposition and Its Applications to Social Network Analysis The Canonical Tensor Decomposition and Its Applications to Social Network Analysis Evrim Acar, Tamara G. Kolda and Daniel M. Dunlavy Sandia National Labs Sandia is a multiprogram laboratory operated by

More information

Elaine T. Hale, Wotao Yin, Yin Zhang

Elaine T. Hale, Wotao Yin, Yin Zhang , Wotao Yin, Yin Zhang Department of Computational and Applied Mathematics Rice University McMaster University, ICCOPT II-MOPTA 2007 August 13, 2007 1 with Noise 2 3 4 1 with Noise 2 3 4 1 with Noise 2

More information

Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images!

Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images! Sparse Solutions of Linear Systems of Equations and Sparse Modeling of Signals and Images! Alfredo Nava-Tudela John J. Benedetto, advisor 1 Happy birthday Lucía! 2 Outline - Problem: Find sparse solutions

More information

CVPR A New Tensor Algebra - Tutorial. July 26, 2017

CVPR A New Tensor Algebra - Tutorial. July 26, 2017 CVPR 2017 A New Tensor Algebra - Tutorial Lior Horesh lhoresh@us.ibm.com Misha Kilmer misha.kilmer@tufts.edu July 26, 2017 Outline Motivation Background and notation New t-product and associated algebraic

More information

Linear Algebra Section 2.6 : LU Decomposition Section 2.7 : Permutations and transposes Wednesday, February 13th Math 301 Week #4

Linear Algebra Section 2.6 : LU Decomposition Section 2.7 : Permutations and transposes Wednesday, February 13th Math 301 Week #4 Linear Algebra Section. : LU Decomposition Section. : Permutations and transposes Wednesday, February 1th Math 01 Week # 1 The LU Decomposition We learned last time that we can factor a invertible matrix

More information

Bindel, Fall 2016 Matrix Computations (CS 6210) Notes for

Bindel, Fall 2016 Matrix Computations (CS 6210) Notes for 1 Logistics Notes for 2016-08-26 1. Our enrollment is at 50, and there are still a few students who want to get in. We only have 50 seats in the room, and I cannot increase the cap further. So if you are

More information

17 Random Projections and Orthogonal Matching Pursuit

17 Random Projections and Orthogonal Matching Pursuit 17 Random Projections and Orthogonal Matching Pursuit Again we will consider high-dimensional data P. Now we will consider the uses and effects of randomness. We will use it to simplify P (put it in a

More information

Sparse BLAS-3 Reduction

Sparse BLAS-3 Reduction Sparse BLAS-3 Reduction to Banded Upper Triangular (Spar3Bnd) Gary Howell, HPC/OIT NC State University gary howell@ncsu.edu Sparse BLAS-3 Reduction p.1/27 Acknowledgements James Demmel, Gene Golub, Franc

More information

A randomized algorithm for approximating the SVD of a matrix

A randomized algorithm for approximating the SVD of a matrix A randomized algorithm for approximating the SVD of a matrix Joint work with Per-Gunnar Martinsson (U. of Colorado) and Vladimir Rokhlin (Yale) Mark Tygert Program in Applied Mathematics Yale University

More information

Robust Principal Component Analysis

Robust Principal Component Analysis ELE 538B: Mathematics of High-Dimensional Data Robust Principal Component Analysis Yuxin Chen Princeton University, Fall 2018 Disentangling sparse and low-rank matrices Suppose we are given a matrix M

More information

Uncorrelated Multilinear Principal Component Analysis through Successive Variance Maximization

Uncorrelated Multilinear Principal Component Analysis through Successive Variance Maximization Uncorrelated Multilinear Principal Component Analysis through Successive Variance Maximization Haiping Lu 1 K. N. Plataniotis 1 A. N. Venetsanopoulos 1,2 1 Department of Electrical & Computer Engineering,

More information

From Matrix to Tensor. Charles F. Van Loan

From Matrix to Tensor. Charles F. Van Loan From Matrix to Tensor Charles F. Van Loan Department of Computer Science January 28, 2016 From Matrix to Tensor From Tensor To Matrix 1 / 68 What is a Tensor? Instead of just A(i, j) it s A(i, j, k) or

More information

arxiv: v1 [cs.it] 21 Feb 2013

arxiv: v1 [cs.it] 21 Feb 2013 q-ary Compressive Sensing arxiv:30.568v [cs.it] Feb 03 Youssef Mroueh,, Lorenzo Rosasco, CBCL, CSAIL, Massachusetts Institute of Technology LCSL, Istituto Italiano di Tecnologia and IIT@MIT lab, Istituto

More information

This work has been submitted to ChesterRep the University of Chester s online research repository.

This work has been submitted to ChesterRep the University of Chester s online research repository. This work has been submitted to ChesterRep the University of Chester s online research repository http://chesterrep.openrepository.com Author(s): Daniel Tock Title: Tensor decomposition and its applications

More information

4.8 Arnoldi Iteration, Krylov Subspaces and GMRES

4.8 Arnoldi Iteration, Krylov Subspaces and GMRES 48 Arnoldi Iteration, Krylov Subspaces and GMRES We start with the problem of using a similarity transformation to convert an n n matrix A to upper Hessenberg form H, ie, A = QHQ, (30) with an appropriate

More information

CS60021: Scalable Data Mining. Dimensionality Reduction

CS60021: Scalable Data Mining. Dimensionality Reduction J. Leskovec, A. Rajaraman, J. Ullman: Mining of Massive Datasets, http://www.mmds.org 1 CS60021: Scalable Data Mining Dimensionality Reduction Sourangshu Bhattacharya Assumption: Data lies on or near a

More information

5.6. PSEUDOINVERSES 101. A H w.

5.6. PSEUDOINVERSES 101. A H w. 5.6. PSEUDOINVERSES 0 Corollary 5.6.4. If A is a matrix such that A H A is invertible, then the least-squares solution to Av = w is v = A H A ) A H w. The matrix A H A ) A H is the left inverse of A and

More information

Randomized Numerical Linear Algebra: Review and Progresses

Randomized Numerical Linear Algebra: Review and Progresses ized ized SVD ized : Review and Progresses Zhihua Department of Computer Science and Engineering Shanghai Jiao Tong University The 12th China Workshop on Machine Learning and Applications Xi an, November

More information

7 Principal Component Analysis

7 Principal Component Analysis 7 Principal Component Analysis This topic will build a series of techniques to deal with high-dimensional data. Unlike regression problems, our goal is not to predict a value (the y-coordinate), it is

More information

DS-GA 1002 Lecture notes 10 November 23, Linear models

DS-GA 1002 Lecture notes 10 November 23, Linear models DS-GA 2 Lecture notes November 23, 2 Linear functions Linear models A linear model encodes the assumption that two quantities are linearly related. Mathematically, this is characterized using linear functions.

More information

Least Squares Approximation

Least Squares Approximation Chapter 6 Least Squares Approximation As we saw in Chapter 5 we can interpret radial basis function interpolation as a constrained optimization problem. We now take this point of view again, but start

More information

Lecture 4. CP and KSVD Representations. Charles F. Van Loan

Lecture 4. CP and KSVD Representations. Charles F. Van Loan Structured Matrix Computations from Structured Tensors Lecture 4. CP and KSVD Representations Charles F. Van Loan Cornell University CIME-EMS Summer School June 22-26, 2015 Cetraro, Italy Structured Matrix

More information

TENSOR LAYERS FOR COMPRESSION OF DEEP LEARNING NETWORKS. Cris Cecka Senior Research Scientist, NVIDIA GTC 2018

TENSOR LAYERS FOR COMPRESSION OF DEEP LEARNING NETWORKS. Cris Cecka Senior Research Scientist, NVIDIA GTC 2018 TENSOR LAYERS FOR COMPRESSION OF DEEP LEARNING NETWORKS Cris Cecka Senior Research Scientist, NVIDIA GTC 2018 Tensors Computations and the GPU AGENDA Tensor Networks and Decompositions Tensor Layers in

More information

PCA, Kernel PCA, ICA

PCA, Kernel PCA, ICA PCA, Kernel PCA, ICA Learning Representations. Dimensionality Reduction. Maria-Florina Balcan 04/08/2015 Big & High-Dimensional Data High-Dimensions = Lot of Features Document classification Features per

More information

EECS 275 Matrix Computation

EECS 275 Matrix Computation EECS 275 Matrix Computation Ming-Hsuan Yang Electrical Engineering and Computer Science University of California at Merced Merced, CA 95344 http://faculty.ucmerced.edu/mhyang Lecture 22 1 / 21 Overview

More information

Computational Linear Algebra

Computational Linear Algebra Computational Linear Algebra PD Dr. rer. nat. habil. Ralf Peter Mundani Computation in Engineering / BGU Scientific Computing in Computer Science / INF Winter Term 2017/18 Part 2: Direct Methods PD Dr.

More information

Singular Value Decomposition and its. SVD and its Applications in Computer Vision

Singular Value Decomposition and its. SVD and its Applications in Computer Vision Singular Value Decomposition and its Applications in Computer Vision Subhashis Banerjee Department of Computer Science and Engineering IIT Delhi October 24, 2013 Overview Linear algebra basics Singular

More information

The multiple-vector tensor-vector product

The multiple-vector tensor-vector product I TD MTVP C KU Leuven August 29, 2013 In collaboration with: N Vanbaelen, K Meerbergen, and R Vandebril Overview I TD MTVP C 1 Introduction Inspiring example Notation 2 Tensor decompositions The CP decomposition

More information

to be more efficient on enormous scale, in a stream, or in distributed settings.

to be more efficient on enormous scale, in a stream, or in distributed settings. 16 Matrix Sketching The singular value decomposition (SVD) can be interpreted as finding the most dominant directions in an (n d) matrix A (or n points in R d ). Typically n > d. It is typically easy to

More information