Single- and multi-variate tensor spectral analysis

1. A method of tensor spectral analysis, comprising:

obtaining a set of Fourier domain singular value decomposition (SVD) factors of a first tensor representing data in multiple dimensions, wherein:

the data corresponds to one or more application-specific variates;

the multiple dimensions include a dimension that represents a measurement; and

the obtaining of the set of Fourier domain SVD factors includes:

performing Fourier transform on the first tensor along a first dimension to obtain a Fourier domain tensor;

performing SVD on the Fourier domain tensor to obtain a first set of Fourier domain SVD tensors;

performing inverse Fourier transform on at least some of the first set of Fourier domain SVD tensors to obtain a set of native domain tensors;

performing Fourier transform on at least some of the set of native domain tensors to obtain a second set of Fourier domain tensors; and

performing SVD on at least some of the second set of Fourier domain tensors to obtain a second set of Fourier domain SVD tensors;

obtaining an SVD result based at least in part on the set of Fourier domain SVD factors of the first tensor, wherein the SVD result indicates energy content of the first tensor; and

outputting at least a portion of the SVD result to be displayed, stored, further processed, or any combination thereof.

2. The method of claim 1 , wherein at least some of the Fourier transform, the SVD, and/or the inverse Fourier transform are performed on a plurality of processors in parallel.

3. The method of claim 1 , further comprising:

reorienting matrices in at least some of the first set of Fourier domain SVD tensors; and

reorienting matrices in at least some of the second set of Fourier domain tensors to provide the set of Fourier domain SVD factors.

4. The method of claim 3 , wherein:

the first set of Fourier domain SVD tensors includes a first Fourier domain left singular vectors tensor, a Fourier domain singular values tensor, and a first Fourier domain right singular vectors tensor; and

the reorienting of matrices in at least some of the first set of Fourier domain SVD tensors includes reorienting matrices in the first Fourier domain left singular vectors tensor according to dominance of singular vectors in the first Fourier domain left singular vectors tensor, and reorienting matrices in the first Fourier domain right singular vectors tensor according to dominance of singular vectors in the first Fourier domain right singular vectors tensor.

5. The method of claim 4 , wherein the reorienting of the matrices in at least some of the second set of Fourier domain tensors includes reorienting:

matrices in a Fourier domain left-left singular vectors tensor according to dominance of singular vectors in the Fourier domain left-left singular vectors tensor;

matrices in a Fourier domain left-right singular vectors tensor according to dominance of singular vectors in the Fourier domain left-right singular vectors tensor;

matrices in a Fourier domain right-left singular vectors tensor according to dominance of singular vectors in the Fourier domain right-left singular vectors tensor; and

matrices in a Fourier domain right-right singular vectors tensor according to dominance of singular vectors in the Fourier domain right-right singular vectors tensor.

6. The method of claim 5 , wherein the set of Fourier domain SVD factors of the first tensor includes a reoriented Fourier domain left-left singular vectors tensor, a Fourier domain left singular values tensor, a reoriented Fourier domain left-right singular vectors tensor, a reoriented Fourier domain right-left singular vectors tensor, a Fourier domain right singular values tensor, a reoriented Fourier domain right-right singular vectors tensor, and the Fourier domain singular values tensor.

7. The method of claim 3 , further comprising performing inverse Fourier transform on:

a reoriented Fourier domain left-left singular vectors tensor to obtain a native domain left-left singular vectors tensor;

a Fourier domain left singular values tensor to obtain a native domain left singular values tensor;

a reoriented Fourier domain left-right singular vectors tensor to obtain a native domain left-right singular vectors tensor;

a reoriented Fourier domain right-left singular vectors tensor to obtain a native domain right-left singular vectors tensor;

a Fourier domain right singular values tensor to obtain a native domain right singular values tensor; and

a reoriented Fourier domain right-right singular vectors tensor to obtain a native domain right-right singular vectors tensor.

8. The method of claim 7 , wherein the SVD result of the first tensor includes:

at least a portion of the native domain left-left singular vectors tensor;

at least a portion of the native domain left singular values tensor;

at least a portion of the native domain left-right singular vectors tensor;

at least a portion of the native domain right-left singular vectors tensor;

at least a portion of the native domain right singular values tensor; and

at least a portion of the native domain right-right singular vectors tensor.

9. The method of claim 1 , wherein the first set of Fourier domain SVD tensors includes a first Fourier domain left singular vectors tensor, a Fourier domain singular values tensor, and a first Fourier domain right singular vectors tensor.

10. The method of claim 9 , wherein the performing of the inverse Fourier transform on the at least some of the first set of Fourier domain SVD tensors includes performing inverse Fourier transform on:

the first Fourier domain left singular vectors tensor along the first dimension to obtain a native domain left singular vectors tensor;

the first Fourier domain right singular vectors tensor along the first dimension to obtain a native domain right singular vectors tensor; and

the Fourier domain singular values tensor along the first dimension to obtain a native domain singular values tensor.

11. The method of claim 10 , wherein the performing of the Fourier transform on the at least some of the set of native domain tensors to obtain the second set of Fourier domain tensors includes performing Fourier transform on:

the native domain left singular vectors tensor along a second dimension to obtain a second Fourier domain left singular vectors tensor; and

the native domain right singular vectors tensor along a third dimension to obtain a second Fourier domain right singular vectors tensor.

12. The method of claim 11 , wherein the performing of SVD on at least some of the second set of Fourier domain tensors to obtain the second set of Fourier domain SVD tensors includes performing SVD on:

the second Fourier domain left singular vectors tensor to obtain a Fourier domain left-left singular vectors tensor, a Fourier domain left singular values tensor, and a Fourier domain left- right singular vectors tensor; and

the second Fourier domain right singular vectors tensor to obtain a Fourier domain right- left singular vectors tensor, a Fourier domain right singular values tensor, and a Fourier domain right-right singular vectors tensor.

13. The method of claim 1 , wherein the at least portion of the SVD result represents a compressed form of the first tensor or a compressed form of a higher order tensor that includes the first tensor.

14. The method of claim 1 , wherein the SVD result that is obtained is the SVD result of the first tensor.

15. The method of claim 1 , wherein the obtaining of the SVD result based at least in part on the set of Fourier domain SVD factors of the first tensor includes performing inverse Fourier transform on a second set of reoriented Fourier domain SVD tensors.

16. The method of claim 1 , wherein:

the first tensor is a 3-dimensional tensor that is included in a higher dimensional tensor that is four-dimensional or higher;

the higher dimensional tensor is of four or more dimensions, and includes one or more additional 3-dimensional tensors; and

the SVD result that is obtained is an SVD result of the higher dimensional tensor.

17. The method of claim 16 , further comprising:

obtaining one or more additional sets of Fourier domain SVD factors of the one or more additional 3-dimensional tensors;

performing Fourier transforms on vectors tensors in the set of Fourier domain SVD factors of the first tensor and the one or more additional sets of Fourier domain SVD factors of the one or more additional 3-dimensional tensors, along a fourth dimension, to obtain corresponding Fourier domain vectors tensors;

performing SVD on the corresponding Fourier domain vectors tensors to obtain corresponding Fourier domain left singular vectors tensors, Fourier domain right singular vectors tensors, and Fourier domain singular values tensors;

reorienting the Fourier domain left singular vectors tensors, Fourier domain right singular vectors tensors, and Fourier domain singular values tensors;

performing inverse Fourier transform on:

the reoriented Fourier domain left singular vectors tensors to obtain native domain left singular vectors tensors;

the reoriented Fourier domain right singular vectors tensors to obtain native domain right singular vectors tensors; and

the reoriented Fourier domain singular values tensors to obtain native domain singular values tensors; and

obtaining the SVD result of the higher dimensional tensor based at least in part on the native domain left singular vectors tensors, the native domain right singular vectors tensors, and the native domain singular values tensors.

18. The method of claim 1 , wherein one or more dominant terms in the SVD result are output and remaining terms are omitted.

19. A system of tensor spectral analysis, comprising:

one or more processors configured to:

obtain a set of Fourier domain singular value decomposition (SVD) factors of a first tensor representing data in multiple dimensions, wherein:

the data corresponds to one or more application-specific variates;

the multiple dimensions include a dimension that represents a measurement and

to obtain the set of Fourier domain SVD factors includes to:

perform Fourier transform on the first tensor along a first dimension to obtain a Fourier domain tensor;

perform SVD on the Fourier domain tensor to obtain a first set of Fourier domain SVD tensors;

perform inverse Fourier transform on at least some of the first set of Fourier domain SVD tensors to obtain a set of native domain tensors;

perform Fourier transform on at least some of the set of native domain tensors to obtain a second set of Fourier domain tensors; and

perform SVD on at least some of the second set of Fourier domain tensors to obtain a second set of Fourier domain SVD tensors;

obtain an SVD result based at least in part on the set of Fourier domain SVD factors of the first tensor, wherein the SVD result indicates energy content of the first tensor; and

output at least a portion of the SVD result to be displayed, stored, further processed, or any combination thereof; and

one or more memories coupled to the one or more processors and configured to provide the one or more processors with instructions.

20. The system of claim 19 , wherein the system comprises a plurality of processors, and at least some of the Fourier transform, the SVD, and/or the inverse Fourier transform are performed on the plurality of processors in parallel.

21. A computer program product for tensor spectral analysis, the computer program product being embodied in a tangible computer readable storage medium and comprising computer instructions for:

obtaining a set of Fourier domain singular value decomposition (SVD) factors of a first tensor representing data in multiple dimensions, wherein:

the data corresponds to one or more application-specific variates;

the multiple dimensions include a dimension that represents a measurement and

the obtaining of the set of Fourier domain SVD factors includes:

performing Fourier transform on the first tensor along a first dimension to obtain a Fourier domain tensor;

performing SVD on the Fourier domain tensor to obtain a first set of Fourier domain SVD tensors;

performing inverse Fourier transform on at least some of the first set of Fourier domain SVD tensors to obtain a set of native domain tensors;

performing Fourier transform on at least some of the set of native domain tensors to obtain a second set of Fourier domain tensors; and

performing SVD on at least some of the second set of Fourier domain tensors to obtain a second set of Fourier domain SVD tensors;

obtaining an SVD result based at least in part on the set of Fourier domain SVD factors of the first tensor, wherein the SVD result indicates energy content of the first tensor; and

outputting at least a portion of the SVD result to be displayed, stored, further processed, or any combination thereof.

22. The computer program product of claim 21 , wherein at least some of the Fourier transform, the SVD, and/or the inverse Fourier transform are performed on a plurality of processors in parallel.