1 / 34

The Power of Selective Memory

The Power of Selective Memory. Shai Shalev-Shwartz Joint work with Ofer Dekel, Yoram Singer Hebrew University, Jerusalem. Outline. Online learning, loss bounds etc. Hypotheses space – PST Margin of prediction and hinge-loss An online learning algorithm Trading margin for depth of the PST

zarifa
Télécharger la présentation

The Power of Selective Memory

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. The Power of Selective Memory Shai Shalev-Shwartz Joint work with Ofer Dekel, Yoram Singer Hebrew University, Jerusalem

  2. Outline • Online learning, loss bounds etc. • Hypotheses space – PST • Margin of prediction and hinge-loss • An online learning algorithm • Trading margin for depth of the PST • Automatic calibration • A self-bounded online algorithm for learning PSTs

  3. Online Learning • For • Get an instance • Predict a target based on • Get true update and suffer loss • Update prediction mechanism

  4. Analysis of Online Algorithm • Relative loss bounds (external regret): For any fixed hypothesis h :

  5. Prediction Suffix Tree (PST) Each hypothesis is parameterized by a triplet: context function

  6. PST Example 0 -3 -1 1 4 -2 7

  7. Margin of Prediction • Margin of prediction • Hinge loss

  8. Complexity of hypothesis • Define the complexity of hypothesis as • We can also extend g s.t. and get

  9. Algorithm I :Learning Unbounded-Depth PST • Init: • For t=1,2,… • Get and predict • Get and suffer loss • Set • Update weight vector • Update tree

  10. Example y = ? y = 0

  11. Example y = ? y = + 0

  12. Example y = ? ? y = + 0

  13. Example y = ? ? y = + - 0 + -.23

  14. Example y = ? ? ? y = + - 0 + -.23

  15. Example y = ? ? ? y = + - + 0 + - .23 -.23 + .16

  16. Example y = ? ? ? - y = + - + 0 + - .23 -.23 + .16

  17. Example y = ? ? ? - y = + - + - 0 + - .23 -.42 + - .16 -.14 + -.09

  18. Example y = ? ? ? - + y = + - + - 0 + - .23 -.42 + - .16 -.14 + -.09

  19. Example y = ? ? ? - + y = + - + - + 0 + - .41 -.42 + - .29 -.14 - + .09 -.09 + .06

  20. Analysis • Let be a sequence of examples and assume that • Let be an arbitrary hypothesis • Let be the loss of on the sequence of examples. Then,

  21. Proof Sketch • Define • Upper bound • Lower bound • Upper + lower bounds give the bound in the theorem

  22. Proof Sketch (Cont.) Where does the lower bound come from? • For simplicity, assume that and • Define a Hilbert space: • The context function gt+1is the projection of gtonto the half-space where f is the function

  23. Example revisited y = + - + - + - + - • The following hypothesis has cumulative loss of 2 and complexity of 2. Therefore, the number of mistakes is bounded above by 12.

  24. Example revisited y = + - + - + - + - • The following hypothesis has cumulative loss of 1 and complexity of 4. Therefore, the number of mistakes is bounded above by 18.But, this tree is very shallow 0 + - 1.41 -1.41 Problem: The tree we learned is much more deeper !

  25. Geometric Intuition

  26. Geometric Intuition (Cont.) Lets force gt+1 to be sparse by “canceling” the new coordinate

  27. Geometric Intuition (Cont.) Now we can show that:

  28. Trading margin for sparsity • We got that • If is much smaller than we can get a loss bound ! • Problem: What happens if is very small and therefore ?Solution: Tolerate small margin errors ! • Conclusion: If we tolerate small margin errors, we can get a sparser tree

  29. Automatic Calibration • Problem: The value of is unknown • Solution: Use the data itself to estimate it ! More specifically: • Denote • If we keep then we get a mistake bound

  30. Algorithm II :Learning Self Bounded-Depth PST • Init: • For t=1,2,… • Get and predict • Get and suffer loss • If do nothing! Otherwise: • Set • Set • Set • Update w and the tree as in Algo. I, up to depth dt

  31. Analysis – Loss Bound • Let be a sequence of examples and assume that • Let be an arbitrary hypothesis • Let be the loss of on the sequence of examples. Then,

  32. Analysis – Bounded depth • Under the previous conditions, the depth of all the trees learned by the algorithm is bounded above by

  33. Performance of Algo. II y = + - + - + - + - … Only 3 mistakes The last PST is of depth 5 The margin is 0.61 (after normalization) The margin of the max margin tree (of infinite depth) is 0.7071 Example revisited 0 - + .55 -.55 + - -. 22 .39 - + .07 -.07 + - .05 -.05 - .03

  34. Conclusions • Discriminative online learning of PSTs • Loss bound • Trade margin and sparsity • Automatic calibration Future work • Experiments • Features selection and extraction • Support vectors selection

More Related