1 of 51

1

Grid Cells: Global positioning system that helps people navigate

2 of 51

2

3 of 51

3

Please Read!

4 of 51

4

Structural and Functional Brain Networks

5 of 51

5

Inside the Nucleus: Chromosome Territories

Tan, Longzhi, et al. "Three-dimensional genome structures of single diploid human cells." Science 361.6405 (2018): 924-928.

6 of 51

6

A Tensor is an 𝑑-way array

 

 

 

 

7 of 51

7

Notations

8 of 51

8

Tensor Unfolding

9 of 51

9

A Common Framework for Tensor Computations

Tensor Unfolding

10 of 51

10

Illustration of row-wise and column-wise unfolding (flattening, matricizing) of a third-order tensor

Andrzej Cichocki, Rafal Zdunek, and Shun'ichi Amari. Nonnegative Matrix and Tensor Factorizations: Applications to Exploratory Multi-way Data Analysis and Blind Source (2009)

11 of 51

11

Unfolding (matricizing) of a third-order tensor

12 of 51

12

SVD Expansion

13 of 51

Rank One Tensors

Building block for decomposition: Vector outer products = Rank one tensors

Matrix version (2 – way)

Rank one matrix

Tensor version (3 – way)

Rank one tensor

14 of 51

Building block for decomposition: Vector outer products = Rank one tensors

 

Visualization

gets weird.. Math is still good!

Rank one Tensors

15 of 51

15

Data

Low-rank model

Matrix notation

Sum of squared errors

Low Rank Approximation

16 of 51

Tensor Decompositions

CP Model: Sum of d-way outer products, useful for interpretation

Tucker Model: Project onto high-variance subspaces to reduce dimensionality

17 of 51

CP Decomposition

The full name is CANDECOMP/PARAFAC Decomposition

CANDECOMP = Canonical Decomposition

PARAFAC = Parallel Factors Decomposition

CP Tensor Factorization (3-way): Detecting low-rank 3-way structure

Higher-order analogue of the SVD

CP first invented in 1927

F. L. Hitchcock, The Expression of a Tensor or a Polyadic as a Sum of Products, Journal of Mathematics and Physics, 1927

18 of 51

Tensor notation

Sum of squared errors

Factor matrices

Component

CP Decomposition

19 of 51

Example 1: Neuronal Activity

Williams, A. H., Kim, T. H., Wang, F., Vyas, S., Ryu, S. I., Shenoy, K. V., ... & Ganguli, S. (2018). Unsupervised discovery of demixed, low-dimensional neural dynamics across multiple timescales through tensor component analysis. Neuron, 98(6), 1099-1115.

20 of 51

Example 1: Neuronal activity

Mouse

in “maze”

Neural activity

One Column of Neuron x

Time matrix

 

× 600 trials (over 5 days)

Microscope by

Inscopix

21 of 51

Trials vary start position and strategies

  • 600 Trials over 5 days
  • Start West or East
  • Conditions swap twice
    • Always turn South
    • Always turn Right
    • Always turn South

Example: Neuronal activity

22 of 51

8-Component CP decomposition of data

23 of 51

8-Component CP decomposition of data

24 of 51

Got the treat! Happy neurons….

8-Component CP decomposition of data

25 of 51

8-Component CP decomposition of data

Clear separation. Starting position Neurons understand mouse has started. ….

26 of 51

8-Component CP decomposition of data

27 of 51

Example 2: Application to hazardous gas

 

Vergara, A., Fonollosa, J., Mahiques, J., Trincavelli, M., Rulkov, N., & Huerta, R. (2013). On the performance of gas sensor arrays in open sampling systems using Inhibitory Support Vector Machines. Sensors and Actuators B: Chemical, 185, 462-477.

28 of 51

Factors

Example 2: Application to hazardous gas

29 of 51

29

Some Mathematics….

30 of 51

30

The Khatri-Rao Product

Definition:

C.R. Rao and S.K. Mitra (1971). Generalized Inverse of Matrices and Applications, John Wiley and Sons, New York

31 of 51

31

The Khatri-Rao Product

If

then the Khatri-Rao product of B and C is given by

Note:

32 of 51

32

The Kronecker Product

33 of 51

33

The Kronecker Product

34 of 51

34

All Possible Entry- Entry Products

35 of 51

35

Kronecker Products of Kronecker Products

36 of 51

36

You have seen them before

37 of 51

37

Block matrix

38 of 51

38

Block Matrices: Addition

39 of 51

39

Block Matrix: Multiplication

40 of 51

40

Block Matrix: Transposition

41 of 51

41

A block matrix can have block matrix entries

42 of 51

42

Reshape

43 of 51

43

Reshape

44 of 51

44

SVD

45 of 51

45

SVD

46 of 51

46

The Tucker product

Definition:

47 of 51

More References

Higher-order SVD

De Lathauwer L, De Moor B, Vandewalle J. A multilinear singular value decomposition. SIAM journal on Matrix Analysis and Applications. 2000;21(4):1253-78.

Tensor SVD

Brazell, M., Li, N., Navasca, C. and Tamon, C., 2013. Solving multilinear systems via tensor inversion. SIAM Journal on Matrix Analysis and Applications, 34(2), pp.542-570.

Tensor-Train

Oseledets, I.V., 2011. Tensor-train decomposition. SIAM Journal on Scientific Computing, 33(5), pp.2295-2317.

Tensor Unfolding

Ragnarsson, S. and Van Loan, C.F., 2012. Block tensor unfoldings. SIAM Journal on Matrix Analysis and Applications, 33(1), pp.149-169.

48 of 51

48

The Tucker product representation

Before…

Now…

49 of 51

49

The Higher-Order SVD (HOSVD)

50 of 51

50

The HOSVD of a matrix

51 of 51

51

The HOSVD of a matrix