1 / 50

Nonnegative Matrix Factorization via Rank-one Downdate

Nonnegative Matrix Factorization via Rank-one Downdate. Nonnegative Matrix Factorization. -2.19. -3.19. -0.02. 1.02. 2 by 1965. 560 by 1965. 560 by 2. 20 by 28. 2 by 1. 2 by 1. 20 by 28. Singular Value Decomposition (SVD). History. History. History. History. History (Algorithms).

maine
Download Presentation

Nonnegative Matrix Factorization via Rank-one Downdate

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Nonnegative Matrix Factorization via Rank-one Downdate

  2. Nonnegative Matrix Factorization

  3. -2.19 -3.19 -0.02 1.02 2 by 1965 560 by 1965 560 by 2 20 by 28 2 by 1 2 by 1 20 by 28

  4. Singular Value Decomposition (SVD)

  5. History

  6. History

  7. History

  8. History

  9. History (Algorithms)

  10. History (Algorithms)

  11. First observation

  12. Power method • Computes the leading singular vectors/value (or eigenvector/value) of a matrix 1 2 while not converged 3 4 5 6 end

  13. Example • Computes the leading singular vectors/value (or eigenvector/value) of a matrix 2 while not converged 3 4 5 6 end

  14. Naive approach to NMF using this observation 1 2 3 4 5 for all set 6 end for • Without step 5, this will simply compute the SVD (Jordan's algorithm, Camille Jordan 1874. )

  15. Second observation

  16. Modified power iteration 1 2 while not converged 3 4 5 6 7 8 end

  17. Modified power iteration: Demo Rank-1 submatrix A = Rank-1 submatrix

  18. Modified power iteration: Demo v: 0.14 0.07 0.64 0.41 0.55 Rank-1 submatrix Rank-1 submatrix

  19. Modified power iteration: Demo v: 0.0 0.00.64 0.41 0.55 Rank-1 submatrix Rank-1 submatrix

  20. Modified power iteration: Demo v: 0.0 0.00.64 0.41 0.55 Rank-1 submatrix Rank-1 submatrix

  21. u: Modified power iteration: Demo v: 0.0 0.00.64 0.410.55 0.16 0.21 0.22 0.44 0.74 0.20 Rank-1 submatrix Rank-1 submatrix

  22. u: Modified power iteration: Demo v: 0.0 0.0 0.64 0.41 0.55 0.0 0.0 0.0 0.44 0.74 0.20 Rank-1 submatrix Rank-1 submatrix

  23. u: Modified power iteration: Demo v: 0.0 0.0 0.64 0.41 0.55 0.0 0.0 0.0 0.44 0.74 0.20 Rank-1 submatrix Rank-1 submatrix

  24. u: Modified power iteration: Demo v: 0.0 0.0 0.60 0.28 0.59 0.0 0.0 0.0 0.44 0.74 0.20 Rank-1 submatrix Rank-1 submatrix

  25. u: Modified power iteration: Demo v: 0.0 0.0 0.60 0.28 0.59 0.0 0.0 0.0 0.44 0.74 0.20 Rank-1 submatrix Rank-1 submatrix

  26. u: Modified power iteration: Demo v: 0.0 0.0 0.60 0.28 0.59 0.0 0.0 0.0 0.44 0.74 0.20 Rank-1 submatrix Rank-1 submatrix Zero-out!

  27. Modified power iteration: Demo Rank-1 submatrix Anew =

  28. Rank-one Downdata (R1D) 1 2 3 4 5 end for

  29. Objective function

  30. ApproxRankOneSubmatrix(A)

  31. Rank-one Downdata (R1D)

  32. Rank-one Downdata (R1D)

  33. A simple model for text

  34. Generating a corpus in the model

  35. Theorem about text

  36. LSI

  37. R1D

  38. Information retrieval task TDT corpus (pilot study, v. 1.3, 1997): news articles Identified topics (first few columns of ): Experimental results Topic: president, clinton, house, white Topic: bosnian, serbs, bosnia, serb, nato, sarajevo, air, bihac Topic: haiti, military, aristide, haitian, troops, port, invasion, … Topic: simpson, defense, judge, case, jury, trial, angeles, los, court, … Topic: bill, today, senate, republicans, house, congress, republican, …

  39. Experimental results • For topic: “OJ Simpson trial” (simpson, defense, judge, case, jury, trial)

  40. Theorem about images

  41. Experimental results

  42. LSI

  43. NMF-DIV

  44. R1D

  45. LSI

  46. NMF_DIV

  47. NMF_SC

  48. R1D

More Related