Harmonic Analysis and Geometries of Digital Data Bases

Size: px
Start display at page:

Download "Harmonic Analysis and Geometries of Digital Data Bases"

Transcription

1 Harmonic Analysis and Geometries of Digital Data Bases AMS Session Special Sesson on the Mathematics of Information and Knowledge, Ronald Coifman (Yale) and Matan Gavish (Stanford, Yale) January 14, 2010 () Harmonic Analysis of Data Bases January 14, / 41

2 Local geometry described by a graph Given a dataset X = {x 1,..., x N } with affinity matrix W i,j (aka covariance, kernel, etc) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

3 Local geometry described by a graph Given a dataset X = {x 1,..., x N } with affinity matrix W i,j (aka covariance, kernel, etc) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

4 Local geometry described by a graph Given a dataset X = {x 1,..., x N } with affinity matrix W i,j (aka covariance, kernel, etc) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

5 Local geometry described by a graph Given a dataset X = {x 1,..., x N } with affinity matrix W i,j (aka covariance, kernel, etc) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

6 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

7 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

8 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

9 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

10 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

11 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

12 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

13 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

14 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

15 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

16 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

17 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

18 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

19 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

20 Mission: Harmonic analysis on n-way arrays (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

21 Why organize {rows} and {columns} of a matrix? In the correct coupled intrinstic geometry of {rows} and {cols} we can analyze matrix effectively (below) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

22 Why organize {rows} and {columns} of a matrix? In the correct coupled intrinstic geometry of {rows} and {cols} we can analyze matrix effectively (below) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

23 e.g.: organize {rows},{columns} of potential operator Consider the matrix M i,j = x i y j 1 ({x i } - red, {y j }- blue ) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

24 e.g.: organize {rows},{columns} of potential operator Consider the matrix M i,j = x i y j 1 ({x i } - red, {y j }- blue ) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

25 e.g.: organize {rows},{columns} of potential operator Scramble rows and columns: consider M i,j = x σ(i) y τ(j) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

26 e.g.: organize {rows},{columns} of potential operator Diffusion embedding of graphs on {rows}, {cols} recovers spatial point layout : (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

27 e.g.: organize {rows},{columns} of potential operator Diffusion embedding of graphs on {rows}, {cols} recovers spatial point layout : (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

28 e.g.: organize torus eigenfunctions Now consider ( the matrix ) sin 2πk j N M k,j = ( ) cos 2π (k m) j N k = 1... m k = m m Partition based correlation { } (Coifman, Gavish) l Harmonic l Analysis d of Data Bases January 14, / 41

29 e.g.: organize torus eigenfunctions Now consider ( the matrix ) sin 2πk j N M k,j = ( ) cos 2π (k m) j N k = 1... m k = m m Partition based correlation { } (Coifman, Gavish) l Harmonic l Analysis d of Data Bases January 14, / 41

30 e.g.: organize torus eigenfunctions Now consider ( the matrix ) sin 2πk j N M k,j = ( ) cos 2π (k m) j N k = 1... m k = m m Partition based correlation For mesh K l = 2 l {0,..., 2 l 1 } d and bump ψ define ρ (f, g) = 2 l l 1 k K l T d ( ) f (x) g (x)ψ 2 l x k dx Can show: ( ) ρ e im x, e im x 1 const m m 2 (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

31 e.g.: organize torus eigenfunctions Diffusion embedding of graphs on {rows}, {cols} recovers points and frequencies organization: (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

32 e.g.: organize torus eigenfunctions Diffusion embedding of graphs on {rows}, {cols} recovers points and frequencies organization: (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

33 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

34 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

35 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

36 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

37 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

38 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

39 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

40 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

41 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

42 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

43 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

44 Graph into Partition Tree (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

45 organize torus eigenfunctions, again Now consider the matrix M i,j = { sin (2πk) cos (2πk) k = 1... m k = m m (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

46 organize torus eigenfunctions, again Now consider the matrix M i,j = { sin (2πk) cos (2πk) k = 1... m k = m m (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

47 organize torus eigenfunctions, again Build affinity on points Create partition tree on points Compute partition-based affinity between eigenvectors (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

48 organize torus eigenfunctions, again Build affinity on points Create partition tree on points Compute partition-based affinity between eigenvectors (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

49 organize torus eigenfunctions, again Build affinity on points Create partition tree on points Compute partition-based affinity between eigenvectors (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

50 e.g.: organize {rows},{columns} of data matrix Term-document matrix A, where A i,j is relative appearence frequency of word i in document j (Data prepared by Priebe et al, COMPSTAT 2004) We recover coupled geometry: hierarchical structure of words (concepts) and hierarchical structure of documents (contexts) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

51 e.g.: organize {rows},{columns} of data matrix (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

52 e.g.: organize {rows},{columns} of data matrix (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

53 e.g.: Image texture separation By Ali Haddad, Yale To each pixel we associate the 8x8 patch around it and compute log absolute value of the Fourier transform as a list of 64 questions. Colors code the partition in each tree level. (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

54 e.g.: Image texture separation By Ali Haddad, Yale To each pixel we associate the 8x8 patch around it and compute log absolute value of the Fourier transform as a list of 64 questions. Colors code the partition in each tree level. (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

55 e.g.: Image texture separation By Ali Haddad, Yale To each pixel we associate the 8x8 patch around it and compute log absolute value of the Fourier transform as a list of 64 questions. Colors code the partition in each tree level. (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

56 e.g.: Image texture separation By Ali Haddad, Yale To each pixel we associate the 8x8 patch around it and compute log absolute value of the Fourier transform as a list of 64 questions. Colors code the partition in each tree level. (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

57 Questions Basic Questions 1 How to quantify the compatibility of database to proposed coupled geometry? 2 How to exploit a compatible coupled geometry for data analysis? (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

58 Questions Basic Questions 1 How to quantify the compatibility of database to proposed coupled geometry? 2 How to exploit a compatible coupled geometry for data analysis? (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

59 Questions Basic Questions 1 How to quantify the compatibility of database to proposed coupled geometry? 2 How to exploit a compatible coupled geometry for data analysis? (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

60 Enter J. Strömberg Strömberg (1998): tensor product of Haar bases extremely efficient in representing regular f : [0, 1] d R Regular = bounded mixed derivatives Using tensor product of Haar bases in all d dimensions sup x [0,1] d f (x) f, ψ R ψ R (x) < const ε logd 1 R R s.t R >ε Only O ( 1 ε logd 1 ( 1 ε)) coefficients needed ( ) 1 ε (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

61 Enter J. Strömberg Strömberg (1998): tensor product of Haar bases extremely efficient in representing regular f : [0, 1] d R Regular = bounded mixed derivatives Using tensor product of Haar bases in all d dimensions sup x [0,1] d f (x) f, ψ R ψ R (x) < const ε logd 1 R R s.t R >ε Only O ( 1 ε logd 1 ( 1 ε)) coefficients needed ( ) 1 ε (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

62 Enter J. Strömberg Strömberg (1998): tensor product of Haar bases extremely efficient in representing regular f : [0, 1] d R Regular = bounded mixed derivatives Using tensor product of Haar bases in all d dimensions sup x [0,1] d f (x) f, ψ R ψ R (x) < const ε logd 1 R R s.t R >ε Only O ( 1 ε logd 1 ( 1 ε)) coefficients needed ( ) 1 ε (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

63 Enter J. Strömberg Strömberg (1998): tensor product of Haar bases extremely efficient in representing regular f : [0, 1] d R Regular = bounded mixed derivatives Using tensor product of Haar bases in all d dimensions sup x [0,1] d f (x) f, ψ R ψ R (x) < const ε logd 1 R R s.t R >ε Only O ( 1 ε logd 1 ( 1 ε)) coefficients needed ( ) 1 ε (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

64 Enter J. Strömberg Strömberg (1998): tensor product of Haar bases extremely efficient in representing regular f : [0, 1] d R Regular = bounded mixed derivatives Using tensor product of Haar bases in all d dimensions sup x [0,1] d f (x) f, ψ R ψ R (x) < const ε logd 1 R R s.t R >ε Only O ( 1 ε logd 1 ( 1 ε)) coefficients needed ( ) 1 ε (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

65 Answers Basic Questions 1 How to quantify the compatibility of database to proposed coupled geometry? 2 How to exploit a compatible coupled geometry for data analysis? Answers Build Haar-like basis {ψ i } for {f : {rows} R} and {ϕ j }for {f : {cols} R} Expand the matrix as M = i,j M, ψ i ϕ j ψ i ϕ j 1 The coefficients l 1 norm i,j M, ψ i ϕ j quantifies compatibility 2 Process M in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

66 Answers Basic Questions 1 How to quantify the compatibility of database to proposed coupled geometry? 2 How to exploit a compatible coupled geometry for data analysis? Answers Build Haar-like basis {ψ i } for {f : {rows} R} and {ϕ j }for {f : {cols} R} Expand the matrix as M = i,j M, ψ i ϕ j ψ i ϕ j 1 The coefficients l 1 norm i,j M, ψ i ϕ j quantifies compatibility 2 Process M in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

67 Answers Basic Questions 1 How to quantify the compatibility of database to proposed coupled geometry? 2 How to exploit a compatible coupled geometry for data analysis? Answers Build Haar-like basis {ψ i } for {f : {rows} R} and {ϕ j }for {f : {cols} R} Expand the matrix as M = i,j M, ψ i ϕ j ψ i ϕ j 1 The coefficients l 1 norm i,j M, ψ i ϕ j quantifies compatibility 2 Process M in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

68 Answers Basic Questions 1 How to quantify the compatibility of database to proposed coupled geometry? 2 How to exploit a compatible coupled geometry for data analysis? Answers Build Haar-like basis {ψ i } for {f : {rows} R} and {ϕ j }for {f : {cols} R} Expand the matrix as M = i,j M, ψ i ϕ j ψ i ϕ j 1 The coefficients l 1 norm i,j M, ψ i ϕ j quantifies compatibility 2 Process M in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

69 Answers Basic Questions 1 How to quantify the compatibility of database to proposed coupled geometry? 2 How to exploit a compatible coupled geometry for data analysis? Answers Build Haar-like basis {ψ i } for {f : {rows} R} and {ϕ j }for {f : {cols} R} Expand the matrix as M = i,j M, ψ i ϕ j ψ i ϕ j 1 The coefficients l 1 norm i,j M, ψ i ϕ j quantifies compatibility 2 Process M in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

70 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

71 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

72 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

73 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

74 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

75 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

76 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

77 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

78 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

79 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

80 The Haar Basis (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

81 Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

82 Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

83 Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

84 Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

85 Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

86 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

87 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

88 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

89 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

90 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

91 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

92 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

93 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

94 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

95 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

96 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

97 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

98 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

99 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

100 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

101 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

102 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

103 Haar-like bases With B. Nadler, Weizmann (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

104 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

105 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

106 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

107 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

108 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

109 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

110 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

111 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

112 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

113 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

114 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

115 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

116 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

117 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

118 Tensor product of Haar-like bases (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

119 Tensor Haar-like basis function (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

120 An approximation theorem Theorem Fix 0 < p < 2, f : X R and ε > 0. Denote A ε f = f, ψ i ψ i (x) 1 i N with f,ψ i >ε 1 p and R(ψ i ) >ε (i.e. retaining only coefficients which are (i) large, and (ii) correspond to basis functions supported on large folders) (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

121 An approximation theorem Theorem 1 # coefficients retained for A ε f ε 1 N i=1 f, ψ i p. In particular it does not depend on d. 2 Approximation in the mean (roughly): X A ε f f p 1 p ε( 1 p 1 2 ) ( N i=1 f, ψ i p ) 1 p. 3 Uniform pointwise approximation holds outside an exceptional set with E p < N i=1 f, ψ i p (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

122 Data Analysis Scheme 1 Construct affinity of rows according to columns 2 Construct affinity of columns according to rows 3 Construct correponding Haar-like bases and compute i,j M, ψ i ϕ j 4 Repreat until converge 5 Expand the matrix M in tensor Haar-like basis and process in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

123 Data Analysis Scheme 1 Construct affinity of rows according to columns 2 Construct affinity of columns according to rows 3 Construct correponding Haar-like bases and compute i,j M, ψ i ϕ j 4 Repreat until converge 5 Expand the matrix M in tensor Haar-like basis and process in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

124 Data Analysis Scheme 1 Construct affinity of rows according to columns 2 Construct affinity of columns according to rows 3 Construct correponding Haar-like bases and compute i,j M, ψ i ϕ j 4 Repreat until converge 5 Expand the matrix M in tensor Haar-like basis and process in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

125 Data Analysis Scheme 1 Construct affinity of rows according to columns 2 Construct affinity of columns according to rows 3 Construct correponding Haar-like bases and compute i,j M, ψ i ϕ j 4 Repreat until converge 5 Expand the matrix M in tensor Haar-like basis and process in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

126 Data Analysis Scheme 1 Construct affinity of rows according to columns 2 Construct affinity of columns according to rows 3 Construct correponding Haar-like bases and compute i,j M, ψ i ϕ j 4 Repreat until converge 5 Expand the matrix M in tensor Haar-like basis and process in coefficient domain (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

127 Data Analysis Scheme i,j M, ψ i ϕ j plotted over iteration number for the term-document matrix data: Σ <M,ψ i > Iteration # (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

128 Data Analysis Scheme X A εf f fraction of coefficients used and theoretical bound over folder volue cutoff ε e(a)*ε 1/2 l1(a A(ε)) l1(a A(ε)) fraction of coeffs used ε (keep coeffs >ε on normalized support>ε) x 10 4 (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

129 Data Analysis Scheme (left) original term-document matrix. (right) after retaining 15% of the tensor Haar-like coefficients (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

130 Application: Numerical operator compression Here is an extension of ideas by Belkin, Coifman, Rokhlin. Back to the charge distribution (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

131 Application: Numerical operator compression the corresponding potential operator (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

132 Application: Numerical operator compression and its scrambled version, which is our data matrix (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

133 Application: Numerical operator compression Coefficient matrix of scrambled operator in the tensor Haar-like basis is exteremely sparse (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

134 Application: Numerical operator compression Coefficient matrix of scrambled operator in the tensor Haar-like basis is exteremely sparse. Working in coefficient domain allows us to consider only coefficients corresponding to functions of support > ε and obtain a Belkin, Coifman, Rokhlin-type compression scheme that does not assume a-priori knowledge of the operator shape or underlying geometry! (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

135 Application: Numerical operator compression operator norm of the residual and fraction of coefficients used over the threshold ε A A(ε) fraction of coeffs used ε (keep coeffs >ε on normalized support>ε) x 10 7 (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

136 Take Home... Sometimes must consider graphs on both rows and columns Strömberg s theory Partition trees and induced tensor Haar-like bases allow quantitative analysis: - Smoothness of matrix w.r.t graphs - Signal processing on matrix New results for Haar bases on R d conjuctured by experimenting with data (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

137 Take Home... Sometimes must consider graphs on both rows and columns Strömberg s theory Partition trees and induced tensor Haar-like bases allow quantitative analysis: - Smoothness of matrix w.r.t graphs - Signal processing on matrix New results for Haar bases on R d conjuctured by experimenting with data (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

138 Take Home... Sometimes must consider graphs on both rows and columns Strömberg s theory Partition trees and induced tensor Haar-like bases allow quantitative analysis: - Smoothness of matrix w.r.t graphs - Signal processing on matrix New results for Haar bases on R d conjuctured by experimenting with data (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

139 Take Home... Sometimes must consider graphs on both rows and columns Strömberg s theory Partition trees and induced tensor Haar-like bases allow quantitative analysis: - Smoothness of matrix w.r.t graphs - Signal processing on matrix New results for Haar bases on R d conjuctured by experimenting with data (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

140 Take Home... Sometimes must consider graphs on both rows and columns Strömberg s theory Partition trees and induced tensor Haar-like bases allow quantitative analysis: - Smoothness of matrix w.r.t graphs - Signal processing on matrix New results for Haar bases on R d conjuctured by experimenting with data (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

141 References R. Coifman and MG, Harmonic Analysis of Digital Data Bases, to appear in Wavelets: Old and New Perspectives, Springer 2010 B. Nadler, MG and R. Coifman, Adaptive Haar-like wavelet analysis on data: theory and inference algorithms, submitted J. O. Strömberg, (1998) Wavelets in higher dimensions, Documenta Mathematica extra volume ICM-1998 (3), R. Coifman and G. Weiss, Extensions of Hardy spaces and their use in analysis, Bul. Of the AMS, 83, #4, 1977, (Coifman, Gavish) Harmonic Analysis of Data Bases January 14, / 41

Multiscale Wavelets on Trees, Graphs and High Dimensional Data

Multiscale Wavelets on Trees, Graphs and High Dimensional Data Multiscale Wavelets on Trees, Graphs and High Dimensional Data ICML 2010, Haifa Matan Gavish (Weizmann/Stanford) Boaz Nadler (Weizmann) Ronald Coifman (Yale) Boaz Nadler Ronald Coifman Motto... the relationships

More information

Global vs. Multiscale Approaches

Global vs. Multiscale Approaches Harmonic Analysis on Graphs Global vs. Multiscale Approaches Weizmann Institute of Science, Rehovot, Israel July 2011 Joint work with Matan Gavish (WIS/Stanford), Ronald Coifman (Yale), ICML 10' Challenge:

More information

Multiscale bi-harmonic Analysis of Digital Data Bases and Earth moving distances.

Multiscale bi-harmonic Analysis of Digital Data Bases and Earth moving distances. Multiscale bi-harmonic Analysis of Digital Data Bases and Earth moving distances. R. Coifman, Department of Mathematics, program of Applied Mathematics Yale University Joint work with M. Gavish and W.

More information

Conference in Honor of Aline Bonami Orleans, June 2014

Conference in Honor of Aline Bonami Orleans, June 2014 Conference in Honor of Aline Bonami Orleans, June 2014 Harmonic Analysis and functional duality, as a tool for organization of information, and learning. R. Coifman Department of Mathematics, program of

More information

Diffusion/Inference geometries of data features, situational awareness and visualization. Ronald R Coifman Mathematics Yale University

Diffusion/Inference geometries of data features, situational awareness and visualization. Ronald R Coifman Mathematics Yale University Diffusion/Inference geometries of data features, situational awareness and visualization Ronald R Coifman Mathematics Yale University Digital data is generally converted to point clouds in high dimensional

More information

Multiscale Wavelets on Trees, Graphs and High Dimensional Data: Theory and Applications to Semi Supervised Learning

Multiscale Wavelets on Trees, Graphs and High Dimensional Data: Theory and Applications to Semi Supervised Learning Multiscale Wavelets on Trees, Graphs and High Dimensional Data: Theory and Applications to Semi Supervised Learning Matan Gavish 1 Boaz Nadler Weizmann Institute of Science, P.O. Box 26, Rehovot, 76100,

More information

March 13, Paper: R.R. Coifman, S. Lafon, Diffusion maps ([Coifman06]) Seminar: Learning with Graphs, Prof. Hein, Saarland University

March 13, Paper: R.R. Coifman, S. Lafon, Diffusion maps ([Coifman06]) Seminar: Learning with Graphs, Prof. Hein, Saarland University Kernels March 13, 2008 Paper: R.R. Coifman, S. Lafon, maps ([Coifman06]) Seminar: Learning with Graphs, Prof. Hein, Saarland University Kernels Figure: Example Application from [LafonWWW] meaningful geometric

More information

Contribution from: Springer Verlag Berlin Heidelberg 2005 ISBN

Contribution from: Springer Verlag Berlin Heidelberg 2005 ISBN Contribution from: Mathematical Physics Studies Vol. 7 Perspectives in Analysis Essays in Honor of Lennart Carleson s 75th Birthday Michael Benedicks, Peter W. Jones, Stanislav Smirnov (Eds.) Springer

More information

Filtering via a Reference Set. A.Haddad, D. Kushnir, R.R. Coifman Technical Report YALEU/DCS/TR-1441 February 21, 2011

Filtering via a Reference Set. A.Haddad, D. Kushnir, R.R. Coifman Technical Report YALEU/DCS/TR-1441 February 21, 2011 Patch-based de-noising algorithms and patch manifold smoothing have emerged as efficient de-noising methods. This paper provides a new insight on these methods, such as the Non Local Means or the image

More information

On the Phase Transition Phenomenon of Graph Laplacian Eigenfunctions on Trees

On the Phase Transition Phenomenon of Graph Laplacian Eigenfunctions on Trees On the Phase Transition Phenomenon of Graph Laplacian Eigenfunctions on Trees Naoki Saito and Ernest Woei Department of Mathematics University of California Davis, CA 9566 USA Email: saito@math.ucdavis.edu;

More information

THE HIDDEN CONVEXITY OF SPECTRAL CLUSTERING

THE HIDDEN CONVEXITY OF SPECTRAL CLUSTERING THE HIDDEN CONVEXITY OF SPECTRAL CLUSTERING Luis Rademacher, Ohio State University, Computer Science and Engineering. Joint work with Mikhail Belkin and James Voss This talk A new approach to multi-way

More information

Risi Kondor, The University of Chicago. Nedelina Teneva UChicago. Vikas K Garg TTI-C, MIT

Risi Kondor, The University of Chicago. Nedelina Teneva UChicago. Vikas K Garg TTI-C, MIT Risi Kondor, The University of Chicago Nedelina Teneva UChicago Vikas K Garg TTI-C, MIT Risi Kondor, The University of Chicago Nedelina Teneva UChicago Vikas K Garg TTI-C, MIT {(x 1, y 1 ), (x 2, y 2 ),,

More information

Diffusion Geometries, Diffusion Wavelets and Harmonic Analysis of large data sets.

Diffusion Geometries, Diffusion Wavelets and Harmonic Analysis of large data sets. Diffusion Geometries, Diffusion Wavelets and Harmonic Analysis of large data sets. R.R. Coifman, S. Lafon, MM Mathematics Department Program of Applied Mathematics. Yale University Motivations The main

More information

Sparse linear models

Sparse linear models Sparse linear models Optimization-Based Data Analysis http://www.cims.nyu.edu/~cfgranda/pages/obda_spring16 Carlos Fernandez-Granda 2/22/2016 Introduction Linear transforms Frequency representation Short-time

More information

Learning on Graphs and Manifolds. CMPSCI 689 Sridhar Mahadevan U.Mass Amherst

Learning on Graphs and Manifolds. CMPSCI 689 Sridhar Mahadevan U.Mass Amherst Learning on Graphs and Manifolds CMPSCI 689 Sridhar Mahadevan U.Mass Amherst Outline Manifold learning is a relatively new area of machine learning (2000-now). Main idea Model the underlying geometry of

More information

Diffusion Geometries, Global and Multiscale

Diffusion Geometries, Global and Multiscale Diffusion Geometries, Global and Multiscale R.R. Coifman, S. Lafon, MM, J.C. Bremer Jr., A.D. Szlam, P.W. Jones, R.Schul Papers, talks, other materials available at: www.math.yale.edu/~mmm82 Data and functions

More information

FILTERING IN THE FREQUENCY DOMAIN

FILTERING IN THE FREQUENCY DOMAIN 1 FILTERING IN THE FREQUENCY DOMAIN Lecture 4 Spatial Vs Frequency domain 2 Spatial Domain (I) Normal image space Changes in pixel positions correspond to changes in the scene Distances in I correspond

More information

Learning Eigenfunctions: Links with Spectral Clustering and Kernel PCA

Learning Eigenfunctions: Links with Spectral Clustering and Kernel PCA Learning Eigenfunctions: Links with Spectral Clustering and Kernel PCA Yoshua Bengio Pascal Vincent Jean-François Paiement University of Montreal April 2, Snowbird Learning 2003 Learning Modal Structures

More information

CSE 291. Assignment Spectral clustering versus k-means. Out: Wed May 23 Due: Wed Jun 13

CSE 291. Assignment Spectral clustering versus k-means. Out: Wed May 23 Due: Wed Jun 13 CSE 291. Assignment 3 Out: Wed May 23 Due: Wed Jun 13 3.1 Spectral clustering versus k-means Download the rings data set for this problem from the course web site. The data is stored in MATLAB format as

More information

Clustering in kernel embedding spaces and organization of documents

Clustering in kernel embedding spaces and organization of documents Clustering in kernel embedding spaces and organization of documents Stéphane Lafon Collaborators: Raphy Coifman (Yale), Yosi Keller (Yale), Ioannis G. Kevrekidis (Princeton), Ann B. Lee (CMU), Boaz Nadler

More information

Graph Partitioning Using Random Walks

Graph Partitioning Using Random Walks Graph Partitioning Using Random Walks A Convex Optimization Perspective Lorenzo Orecchia Computer Science Why Spectral Algorithms for Graph Problems in practice? Simple to implement Can exploit very efficient

More information

SOS Boosting of Image Denoising Algorithms

SOS Boosting of Image Denoising Algorithms SOS Boosting of Image Denoising Algorithms Yaniv Romano and Michael Elad The Technion Israel Institute of technology Haifa 32000, Israel The research leading to these results has received funding from

More information

Wavelets and Image Compression Augusta State University April, 27, Joe Lakey. Department of Mathematical Sciences. New Mexico State University

Wavelets and Image Compression Augusta State University April, 27, Joe Lakey. Department of Mathematical Sciences. New Mexico State University Wavelets and Image Compression Augusta State University April, 27, 6 Joe Lakey Department of Mathematical Sciences New Mexico State University 1 Signals and Images Goal Reduce image complexity with little

More information

Adaptive Compressive Imaging Using Sparse Hierarchical Learned Dictionaries

Adaptive Compressive Imaging Using Sparse Hierarchical Learned Dictionaries Adaptive Compressive Imaging Using Sparse Hierarchical Learned Dictionaries Jarvis Haupt University of Minnesota Department of Electrical and Computer Engineering Supported by Motivation New Agile Sensing

More information

Diffusion Maps, Spectral Clustering and Eigenfunctions of Fokker-Planck Operators

Diffusion Maps, Spectral Clustering and Eigenfunctions of Fokker-Planck Operators Diffusion Maps, Spectral Clustering and Eigenfunctions of Fokker-Planck Operators Boaz Nadler Stéphane Lafon Ronald R. Coifman Department of Mathematics, Yale University, New Haven, CT 652. {boaz.nadler,stephane.lafon,ronald.coifman}@yale.edu

More information

Fourier Series and Recent Developments in Analysis

Fourier Series and Recent Developments in Analysis Fourier Series and Recent Developments in Analysis Karlstad, June 2003 Javier Soria (U. Barcelona) 1 Jean Baptiste Joseph Fourier (1768-1830) It was around 1804 that Fourier did his important mathematical

More information

Intro to harmonic analysis on groups Risi Kondor

Intro to harmonic analysis on groups Risi Kondor Risi Kondor Any (sufficiently smooth) function f on the unit circle (equivalently, any 2π periodic f ) can be decomposed into a sum of sinusoidal waves f(x) = k= c n e ikx c n = 1 2π f(x) e ikx dx 2π 0

More information

Learning gradients: prescriptive models

Learning gradients: prescriptive models Department of Statistical Science Institute for Genome Sciences & Policy Department of Computer Science Duke University May 11, 2007 Relevant papers Learning Coordinate Covariances via Gradients. Sayan

More information

Estimation of large dimensional sparse covariance matrices

Estimation of large dimensional sparse covariance matrices Estimation of large dimensional sparse covariance matrices Department of Statistics UC, Berkeley May 5, 2009 Sample covariance matrix and its eigenvalues Data: n p matrix X n (independent identically distributed)

More information

A NEW BASIS SELECTION PARADIGM FOR WAVELET PACKET IMAGE CODING

A NEW BASIS SELECTION PARADIGM FOR WAVELET PACKET IMAGE CODING A NEW BASIS SELECTION PARADIGM FOR WAVELET PACKET IMAGE CODING Nasir M. Rajpoot, Roland G. Wilson, François G. Meyer, Ronald R. Coifman Corresponding Author: nasir@dcs.warwick.ac.uk ABSTRACT In this paper,

More information

Analysis of Fractals, Image Compression and Entropy Encoding

Analysis of Fractals, Image Compression and Entropy Encoding Analysis of Fractals, Image Compression and Entropy Encoding Myung-Sin Song Southern Illinois University Edwardsville Jul 10, 2009 Joint work with Palle Jorgensen. Outline 1. Signal and Image processing,

More information

Spectral Graph Theory and its Applications. Daniel A. Spielman Dept. of Computer Science Program in Applied Mathematics Yale Unviersity

Spectral Graph Theory and its Applications. Daniel A. Spielman Dept. of Computer Science Program in Applied Mathematics Yale Unviersity Spectral Graph Theory and its Applications Daniel A. Spielman Dept. of Computer Science Program in Applied Mathematics Yale Unviersity Outline Adjacency matrix and Laplacian Intuition, spectral graph drawing

More information

Detecting Sparse Structures in Data in Sub-Linear Time: A group testing approach

Detecting Sparse Structures in Data in Sub-Linear Time: A group testing approach Detecting Sparse Structures in Data in Sub-Linear Time: A group testing approach Boaz Nadler The Weizmann Institute of Science Israel Joint works with Inbal Horev, Ronen Basri, Meirav Galun and Ery Arias-Castro

More information

Geometry on Probability Spaces

Geometry on Probability Spaces Geometry on Probability Spaces Steve Smale Toyota Technological Institute at Chicago 427 East 60th Street, Chicago, IL 60637, USA E-mail: smale@math.berkeley.edu Ding-Xuan Zhou Department of Mathematics,

More information

Value Function Approximation with Diffusion Wavelets and Laplacian Eigenfunctions

Value Function Approximation with Diffusion Wavelets and Laplacian Eigenfunctions Value Function Approximation with Diffusion Wavelets and Laplacian Eigenfunctions Sridhar Mahadevan Department of Computer Science University of Massachusetts Amherst, MA 13 mahadeva@cs.umass.edu Mauro

More information

Diffusion Wavelets and Applications

Diffusion Wavelets and Applications Diffusion Wavelets and Applications J.C. Bremer, R.R. Coifman, P.W. Jones, S. Lafon, M. Mohlenkamp, MM, R. Schul, A.D. Szlam Demos, web pages and preprints available at: S.Lafon: www.math.yale.edu/~sl349

More information

Fast Hard Thresholding with Nesterov s Gradient Method

Fast Hard Thresholding with Nesterov s Gradient Method Fast Hard Thresholding with Nesterov s Gradient Method Volkan Cevher Idiap Research Institute Ecole Polytechnique Federale de ausanne volkan.cevher@epfl.ch Sina Jafarpour Department of Computer Science

More information

and finally, any second order divergence form elliptic operator

and finally, any second order divergence form elliptic operator Supporting Information: Mathematical proofs Preliminaries Let be an arbitrary bounded open set in R n and let L be any elliptic differential operator associated to a symmetric positive bilinear form B

More information

From graph to manifold Laplacian: The convergence rate

From graph to manifold Laplacian: The convergence rate Appl. Comput. Harmon. Anal. 2 (2006) 28 34 www.elsevier.com/locate/acha Letter to the Editor From graph to manifold Laplacian: The convergence rate A. Singer Department of athematics, Yale University,

More information

Deep Learning: Approximation of Functions by Composition

Deep Learning: Approximation of Functions by Composition Deep Learning: Approximation of Functions by Composition Zuowei Shen Department of Mathematics National University of Singapore Outline 1 A brief introduction of approximation theory 2 Deep learning: approximation

More information

Stable MMPI-2 Scoring: Introduction to Kernel. Extension Techniques

Stable MMPI-2 Scoring: Introduction to Kernel. Extension Techniques 1 Stable MMPI-2 Scoring: Introduction to Kernel Extension Techniques Liberty,E., Almagor,M., Zucker,S., Keller,Y., and Coifman,R.R. Abstract The current study introduces a new technique called Geometric

More information

Bi-stochastic kernels via asymmetric affinity functions

Bi-stochastic kernels via asymmetric affinity functions Bi-stochastic kernels via asymmetric affinity functions Ronald R. Coifman, Matthew J. Hirn Yale University Department of Mathematics P.O. Box 208283 New Haven, Connecticut 06520-8283 USA ariv:1209.0237v4

More information

Multiscale Approach for the Network Compression-friendly Ordering

Multiscale Approach for the Network Compression-friendly Ordering Multiscale Approach for the Network Compression-friendly Ordering Ilya Safro (Argonne National Laboratory) and Boris Temkin (Weizmann Institute of Science) SIAM Parallel Processing for Scientific Computing

More information

Kernels A Machine Learning Overview

Kernels A Machine Learning Overview Kernels A Machine Learning Overview S.V.N. Vishy Vishwanathan vishy@axiom.anu.edu.au National ICT of Australia and Australian National University Thanks to Alex Smola, Stéphane Canu, Mike Jordan and Peter

More information

NORMALIZED CUTS ARE APPROXIMATELY INVERSE EXIT TIMES

NORMALIZED CUTS ARE APPROXIMATELY INVERSE EXIT TIMES NORMALIZED CUTS ARE APPROXIMATELY INVERSE EXIT TIMES MATAN GAVISH AND BOAZ NADLER Abstract The Normalized Cut is a widely used measure of separation between clusters in a graph In this paper we provide

More information

The Generalized Haar-Walsh Transform (GHWT) for Data Analysis on Graphs and Networks

The Generalized Haar-Walsh Transform (GHWT) for Data Analysis on Graphs and Networks The Generalized Haar-Walsh Transform (GHWT) for Data Analysis on Graphs and Networks Jeff Irion & Naoki Saito Department of Mathematics University of California, Davis SIAM Annual Meeting 2014 Chicago,

More information

Pulse characterization with Wavelet transforms combined with classification using binary arrays

Pulse characterization with Wavelet transforms combined with classification using binary arrays Pulse characterization with Wavelet transforms combined with classification using binary arrays Overview Wavelet Transformation Creating binary arrays out of and how to deal with them An estimator for

More information

Satellite image deconvolution using complex wavelet packets

Satellite image deconvolution using complex wavelet packets Satellite image deconvolution using complex wavelet packets André Jalobeanu, Laure Blanc-Féraud, Josiane Zerubia ARIANA research group INRIA Sophia Antipolis, France CNRS / INRIA / UNSA www.inria.fr/ariana

More information

Spectral Hashing: Learning to Leverage 80 Million Images

Spectral Hashing: Learning to Leverage 80 Million Images Spectral Hashing: Learning to Leverage 80 Million Images Yair Weiss, Antonio Torralba, Rob Fergus Hebrew University, MIT, NYU Outline Motivation: Brute Force Computer Vision. Semantic Hashing. Spectral

More information

Statistical Inference on Large Contingency Tables: Convergence, Testability, Stability. COMPSTAT 2010 Paris, August 23, 2010

Statistical Inference on Large Contingency Tables: Convergence, Testability, Stability. COMPSTAT 2010 Paris, August 23, 2010 Statistical Inference on Large Contingency Tables: Convergence, Testability, Stability Marianna Bolla Institute of Mathematics Budapest University of Technology and Economics marib@math.bme.hu COMPSTAT

More information

Multiresolution analysis & wavelets (quick tutorial)

Multiresolution analysis & wavelets (quick tutorial) Multiresolution analysis & wavelets (quick tutorial) Application : image modeling André Jalobeanu Multiresolution analysis Set of closed nested subspaces of j = scale, resolution = 2 -j (dyadic wavelets)

More information

Multiscale Manifold Learning

Multiscale Manifold Learning Multiscale Manifold Learning Chang Wang IBM T J Watson Research Lab Kitchawan Rd Yorktown Heights, New York 598 wangchan@usibmcom Sridhar Mahadevan Computer Science Department University of Massachusetts

More information

Multiresolution analysis on the symmetric group

Multiresolution analysis on the symmetric group Multiresolution analysis on the symmetric group Risi Kondor and Walter Dempsey Department of Statistics and Department of Computer Science The University of Chicago {risi,wdempsey}@uchicago.edu Abstract

More information

Achieving scale covariance

Achieving scale covariance Achieving scale covariance Goal: independently detect corresponding regions in scaled versions of the same image Need scale selection mechanism for finding characteristic region size that is covariant

More information

Fast algorithms for dimensionality reduction and data visualization

Fast algorithms for dimensionality reduction and data visualization Fast algorithms for dimensionality reduction and data visualization Manas Rachh Yale University 1/33 Acknowledgements George Linderman (Yale) Jeremy Hoskins (Yale) Stefan Steinerberger (Yale) Yuval Kluger

More information

Graphs, Geometry and Semi-supervised Learning

Graphs, Geometry and Semi-supervised Learning Graphs, Geometry and Semi-supervised Learning Mikhail Belkin The Ohio State University, Dept of Computer Science and Engineering and Dept of Statistics Collaborators: Partha Niyogi, Vikas Sindhwani In

More information

Continuous Probability Distributions from Finite Data. Abstract

Continuous Probability Distributions from Finite Data. Abstract LA-UR-98-3087 Continuous Probability Distributions from Finite Data David M. Schmidt Biophysics Group, Los Alamos National Laboratory, Los Alamos, New Mexico 87545 (August 5, 1998) Abstract Recent approaches

More information

2. the basis functions have different symmetries. 1 k = 0. x( t) 1 t 0 x(t) 0 t 1

2. the basis functions have different symmetries. 1 k = 0. x( t) 1 t 0 x(t) 0 t 1 In the next few lectures, we will look at a few examples of orthobasis expansions that are used in modern signal processing. Cosine transforms The cosine-i transform is an alternative to Fourier series;

More information

Limits of Spectral Clustering

Limits of Spectral Clustering Limits of Spectral Clustering Ulrike von Luxburg and Olivier Bousquet Max Planck Institute for Biological Cybernetics Spemannstr. 38, 72076 Tübingen, Germany {ulrike.luxburg,olivier.bousquet}@tuebingen.mpg.de

More information

Introduction to Alternating Direction Method of Multipliers

Introduction to Alternating Direction Method of Multipliers Introduction to Alternating Direction Method of Multipliers Yale Chang Machine Learning Group Meeting September 29, 2016 Yale Chang (Machine Learning Group Meeting) Introduction to Alternating Direction

More information

Algorithm S1. Nonlinear Laplacian spectrum analysis (NLSA)

Algorithm S1. Nonlinear Laplacian spectrum analysis (NLSA) Algorithm S1. Nonlinear Laplacian spectrum analysis (NLSA) input : data array x of size m S lag window q Gaussian width ɛ number of nearest neighbors b number of Laplacian eigenfunctions l output: array

More information

sparse and low-rank tensor recovery Cubic-Sketching

sparse and low-rank tensor recovery Cubic-Sketching Sparse and Low-Ran Tensor Recovery via Cubic-Setching Guang Cheng Department of Statistics Purdue University www.science.purdue.edu/bigdata CCAM@Purdue Math Oct. 27, 2017 Joint wor with Botao Hao and Anru

More information

DIFFUSION MAPS, REDUCTION COORDINATES AND LOW DIMENSIONAL REPRESENTATION OF STOCHASTIC SYSTEMS

DIFFUSION MAPS, REDUCTION COORDINATES AND LOW DIMENSIONAL REPRESENTATION OF STOCHASTIC SYSTEMS DIFFUSION MAPS, REDUCTION COORDINATES AND LOW DIMENSIONAL REPRESENTATION OF STOCHASTIC SYSTEMS R.R. COIFMAN, I.G. KEVREKIDIS, S. LAFON, M. MAGGIONI, AND B. NADLER Abstract. The concise representation of

More information

A Multiscale Framework for Markov Decision Processes using Diffusion Wavelets

A Multiscale Framework for Markov Decision Processes using Diffusion Wavelets A Multiscale Framework for Markov Decision Processes using Diffusion Wavelets Mauro Maggioni Program in Applied Mathematics Department of Mathematics Yale University New Haven, CT 6 mauro.maggioni@yale.edu

More information

Karhunen-Loève Approximation of Random Fields Using Hierarchical Matrix Techniques

Karhunen-Loève Approximation of Random Fields Using Hierarchical Matrix Techniques Institut für Numerische Mathematik und Optimierung Karhunen-Loève Approximation of Random Fields Using Hierarchical Matrix Techniques Oliver Ernst Computational Methods with Applications Harrachov, CR,

More information

Diffusion Maps, Spectral Clustering and Eigenfunctions of Fokker-Planck Operators

Diffusion Maps, Spectral Clustering and Eigenfunctions of Fokker-Planck Operators Diffusion Maps, Spectral Clustering and Eigenfunctions of Fokker-Planck Operators Boaz Nadler Stéphane Lafon Ronald R. Coifman Department of Mathematics, Yale University, New Haven, CT 652. {boaz.nadler,stephane.lafon,ronald.coifman}@yale.edu

More information

524 Jan-Olov Stromberg practice, this imposes strong restrictions both on N and on d. The current state of approximation theory is essentially useless

524 Jan-Olov Stromberg practice, this imposes strong restrictions both on N and on d. The current state of approximation theory is essentially useless Doc. Math. J. DMV 523 Computation with Wavelets in Higher Dimensions Jan-Olov Stromberg 1 Abstract. In dimension d, a lattice grid of size N has N d points. The representation of a function by, for instance,

More information

Fast Direct Policy Evaluation using Multiscale Analysis of Markov Diffusion Processes

Fast Direct Policy Evaluation using Multiscale Analysis of Markov Diffusion Processes Fast Direct Policy Evaluation using Multiscale Analysis of Markov Diffusion Processes Mauro Maggioni mauro.maggioni@yale.edu Department of Mathematics, Yale University, P.O. Box 88, New Haven, CT,, U.S.A.

More information

LoG Blob Finding and Scale. Scale Selection. Blobs (and scale selection) Achieving scale covariance. Blob detection in 2D. Blob detection in 2D

LoG Blob Finding and Scale. Scale Selection. Blobs (and scale selection) Achieving scale covariance. Blob detection in 2D. Blob detection in 2D Achieving scale covariance Blobs (and scale selection) Goal: independently detect corresponding regions in scaled versions of the same image Need scale selection mechanism for finding characteristic region

More information

Multiscale Analysis and Diffusion Semigroups With Applications

Multiscale Analysis and Diffusion Semigroups With Applications Multiscale Analysis and Diffusion Semigroups With Applications Karamatou Yacoubou Djima Advisor: Wojciech Czaja Norbert Wiener Center Department of Mathematics University of Maryland, College Park http://www.norbertwiener.umd.edu

More information

Simple Cell Receptive Fields in V1.

Simple Cell Receptive Fields in V1. Simple Cell Receptive Fields in V1. The receptive field properties of simple cells in V1 were studied by Hubel and Wiesel [65][66] who showed that many cells were tuned to the orientation of edges and

More information

Self-Tuning Semantic Image Segmentation

Self-Tuning Semantic Image Segmentation Self-Tuning Semantic Image Segmentation Sergey Milyaev 1,2, Olga Barinova 2 1 Voronezh State University sergey.milyaev@gmail.com 2 Lomonosov Moscow State University obarinova@graphics.cs.msu.su Abstract.

More information

DIMENSION REDUCTION. min. j=1

DIMENSION REDUCTION. min. j=1 DIMENSION REDUCTION 1 Principal Component Analysis (PCA) Principal components analysis (PCA) finds low dimensional approximations to the data by projecting the data onto linear subspaces. Let X R d and

More information

Multiscale analysis on graphs

Multiscale analysis on graphs Mathematics and Computer Science Duke University I.P.A.M. 11/17/08 In collaboration with R.R. Coifman, P.W. Jones, Y-M. Jung, R. Schul, A.D. Szlam & J.C. Bremer Jr. Funding: NSF/DHS-FODAVA, DMS, IIS, CCF;

More information

Beyond Scalar Affinities for Network Analysis or Vector Diffusion Maps and the Connection Laplacian

Beyond Scalar Affinities for Network Analysis or Vector Diffusion Maps and the Connection Laplacian Beyond Scalar Affinities for Network Analysis or Vector Diffusion Maps and the Connection Laplacian Amit Singer Princeton University Department of Mathematics and Program in Applied and Computational Mathematics

More information

Wavelets and Image Compression. Bradley J. Lucier

Wavelets and Image Compression. Bradley J. Lucier Wavelets and Image Compression Bradley J. Lucier Abstract. In this paper we present certain results about the compression of images using wavelets. We concentrate on the simplest case of the Haar decomposition

More information

Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials

Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials Philipp Krähenbühl and Vladlen Koltun Stanford University Presenter: Yuan-Ting Hu 1 Conditional Random Field (CRF) E x I = φ u

More information

Scale-Invariance of Support Vector Machines based on the Triangular Kernel. Abstract

Scale-Invariance of Support Vector Machines based on the Triangular Kernel. Abstract Scale-Invariance of Support Vector Machines based on the Triangular Kernel François Fleuret Hichem Sahbi IMEDIA Research Group INRIA Domaine de Voluceau 78150 Le Chesnay, France Abstract This paper focuses

More information

Applied Machine Learning for Biomedical Engineering. Enrico Grisan

Applied Machine Learning for Biomedical Engineering. Enrico Grisan Applied Machine Learning for Biomedical Engineering Enrico Grisan enrico.grisan@dei.unipd.it Data representation To find a representation that approximates elements of a signal class with a linear combination

More information

Certifying the Global Optimality of Graph Cuts via Semidefinite Programming: A Theoretic Guarantee for Spectral Clustering

Certifying the Global Optimality of Graph Cuts via Semidefinite Programming: A Theoretic Guarantee for Spectral Clustering Certifying the Global Optimality of Graph Cuts via Semidefinite Programming: A Theoretic Guarantee for Spectral Clustering Shuyang Ling Courant Institute of Mathematical Sciences, NYU Aug 13, 2018 Joint

More information

Computation of operators in wavelet coordinates

Computation of operators in wavelet coordinates Computation of operators in wavelet coordinates Tsogtgerel Gantumur and Rob Stevenson Department of Mathematics Utrecht University Tsogtgerel Gantumur - Computation of operators in wavelet coordinates

More information

Sensing systems limited by constraints: physical size, time, cost, energy

Sensing systems limited by constraints: physical size, time, cost, energy Rebecca Willett Sensing systems limited by constraints: physical size, time, cost, energy Reduce the number of measurements needed for reconstruction Higher accuracy data subject to constraints Original

More information

2.3. Clustering or vector quantization 57

2.3. Clustering or vector quantization 57 Multivariate Statistics non-negative matrix factorisation and sparse dictionary learning The PCA decomposition is by construction optimal solution to argmin A R n q,h R q p X AH 2 2 under constraint :

More information

Graph Functional Methods for Climate Partitioning

Graph Functional Methods for Climate Partitioning Graph Functional Methods for Climate Partitioning Mathilde Mougeot - with D. Picard, V. Lefieux*, M. Marchand* Université Paris Diderot, France *Réseau Transport Electrique (RTE) Buenos Aires, 2015 Mathilde

More information

Justin Solomon MIT, Spring 2017

Justin Solomon MIT, Spring 2017 Justin Solomon MIT, Spring 2017 http://pngimg.com/upload/hammer_png3886.png You can learn a lot about a shape by hitting it (lightly) with a hammer! What can you learn about its shape from vibration frequencies

More information

Supervised locally linear embedding

Supervised locally linear embedding Supervised locally linear embedding Dick de Ridder 1, Olga Kouropteva 2, Oleg Okun 2, Matti Pietikäinen 2 and Robert P.W. Duin 1 1 Pattern Recognition Group, Department of Imaging Science and Technology,

More information

Edges and Scale. Image Features. Detecting edges. Origin of Edges. Solution: smooth first. Effects of noise

Edges and Scale. Image Features. Detecting edges. Origin of Edges. Solution: smooth first. Effects of noise Edges and Scale Image Features From Sandlot Science Slides revised from S. Seitz, R. Szeliski, S. Lazebnik, etc. Origin of Edges surface normal discontinuity depth discontinuity surface color discontinuity

More information

A Statistical Look at Spectral Graph Analysis. Deep Mukhopadhyay

A Statistical Look at Spectral Graph Analysis. Deep Mukhopadhyay A Statistical Look at Spectral Graph Analysis Deep Mukhopadhyay Department of Statistics, Temple University Office: Speakman 335 deep@temple.edu http://sites.temple.edu/deepstat/ Graph Signal Processing

More information

Approximate Message Passing

Approximate Message Passing Approximate Message Passing Mohammad Emtiyaz Khan CS, UBC February 8, 2012 Abstract In this note, I summarize Sections 5.1 and 5.2 of Arian Maleki s PhD thesis. 1 Notation We denote scalars by small letters

More information

ECE 901 Lecture 16: Wavelet Approximation Theory

ECE 901 Lecture 16: Wavelet Approximation Theory ECE 91 Lecture 16: Wavelet Approximation Theory R. Nowak 5/17/29 1 Introduction In Lecture 4 and 15, we investigated the problem of denoising a smooth signal in additive white noise. In Lecture 4, we considered

More information

Spectral Graph Wavelets on the Cortical Connectome and Regularization of the EEG Inverse Problem

Spectral Graph Wavelets on the Cortical Connectome and Regularization of the EEG Inverse Problem Spectral Graph Wavelets on the Cortical Connectome and Regularization of the EEG Inverse Problem David K Hammond University of Oregon / NeuroInformatics Center International Conference on Industrial and

More information

Interpolation via weighted l 1 -minimization

Interpolation via weighted l 1 -minimization Interpolation via weighted l 1 -minimization Holger Rauhut RWTH Aachen University Lehrstuhl C für Mathematik (Analysis) Mathematical Analysis and Applications Workshop in honor of Rupert Lasser Helmholtz

More information

Module 4 MULTI- RESOLUTION ANALYSIS. Version 2 ECE IIT, Kharagpur

Module 4 MULTI- RESOLUTION ANALYSIS. Version 2 ECE IIT, Kharagpur Module 4 MULTI- RESOLUTION ANALYSIS Lesson Theory of Wavelets Instructional Objectives At the end of this lesson, the students should be able to:. Explain the space-frequency localization problem in sinusoidal

More information

Wavelets Marialuce Graziadei

Wavelets Marialuce Graziadei Wavelets Marialuce Graziadei 1. A brief summary 2. Vanishing moments 3. 2D-wavelets 4. Compression 5. De-noising 1 1. A brief summary φ(t): scaling function. For φ the 2-scale relation hold φ(t) = p k

More information

Machine Learning: Basis and Wavelet 김화평 (CSE ) Medical Image computing lab 서진근교수연구실 Haar DWT in 2 levels

Machine Learning: Basis and Wavelet 김화평 (CSE ) Medical Image computing lab 서진근교수연구실 Haar DWT in 2 levels Machine Learning: Basis and Wavelet 32 157 146 204 + + + + + - + - 김화평 (CSE ) Medical Image computing lab 서진근교수연구실 7 22 38 191 17 83 188 211 71 167 194 207 135 46 40-17 18 42 20 44 31 7 13-32 + + - - +

More information

Thermodynamic limit for a system of interacting fermions in a random medium. Pieces one-dimensional model

Thermodynamic limit for a system of interacting fermions in a random medium. Pieces one-dimensional model Thermodynamic limit for a system of interacting fermions in a random medium. Pieces one-dimensional model Nikolaj Veniaminov (in collaboration with Frédéric Klopp) CEREMADE, University of Paris IX Dauphine

More information

Module 7:Data Representation Lecture 35: Wavelets. The Lecture Contains: Wavelets. Discrete Wavelet Transform (DWT) Haar wavelets: Example

Module 7:Data Representation Lecture 35: Wavelets. The Lecture Contains: Wavelets. Discrete Wavelet Transform (DWT) Haar wavelets: Example The Lecture Contains: Wavelets Discrete Wavelet Transform (DWT) Haar wavelets: Example Haar wavelets: Theory Matrix form Haar wavelet matrices Dimensionality reduction using Haar wavelets file:///c /Documents%20and%20Settings/iitkrana1/My%20Documents/Google%20Talk%20Received%20Files/ist_data/lecture35/35_1.htm[6/14/2012

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

Universal Scalable Robust Solvers from Computational Information Games and fast eigenspace adapted Multiresolution Analysis.

Universal Scalable Robust Solvers from Computational Information Games and fast eigenspace adapted Multiresolution Analysis. Universal Scalable Robust Solvers from Computational Information Games and fast eigenspace adapted Multiresolution Analysis Houman Owhadi Joint work with Clint Scovel IPAM Apr 3, 2017 DARPA EQUiPS / AFOSR

More information

Exploiting Sparse Non-Linear Structure in Astronomical Data

Exploiting Sparse Non-Linear Structure in Astronomical Data Exploiting Sparse Non-Linear Structure in Astronomical Data Ann B. Lee Department of Statistics and Department of Machine Learning, Carnegie Mellon University Joint work with P. Freeman, C. Schafer, and

More information