1 / 98

AdaBoost

AdaBoost. Classifier. Simplest classifier. Adaboost : Agenda. ( Ada ptive Boost ing, R. Scharpire , Y. Freund, ICML, 1996): Supervised classifier Assembling classifiers Combine many low-accuracy classifiers (weak learners) to create a high-accuracy classifier (strong learners ).

kalona
Download Presentation

AdaBoost

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. AdaBoost

  2. Classifier • Simplest classifier

  3. Adaboost: Agenda • (Adaptive Boosting, R. Scharpire, Y. Freund, ICML, 1996): • Supervised classifier • Assembling classifiers • Combine many low-accuracy classifiers (weak learners) to create a high-accuracy classifier (strong learners)

  4. Example 1

  5. Adaboost: Example (1/10)

  6. Adaboost: Example (2/10)

  7. Adaboost: Example (3/10)

  8. Adaboost: Example (4/10)

  9. Adaboost: Example (5/10)

  10. Adaboost: Example (6/10)

  11. Adaboost: Example (7/10)

  12. Adaboost: Example (8/10)

  13. Adaboost: Example (9/10)

  14. Adaboost: Example (10/10)

  15. Adaboost • Strong classifier = linear combination of T weak classifiers (1) Design of weak classifier (2) Weight for each classifier (Hypothesis weight) (3) Update weight for each data (example distribution) • Weak Classifier: < 50% error over any distribution

  16. Adaboost: Terminology (1/2)

  17. Adaboost: Terminology (2/2)

  18. Adaboost: Framework

  19. Adaboost: Framework

  20. Adaboost • Strong classifier = linear combination of T weak classifiers (1) Design of weak classifier (2) Weight for each classifier (Hypothesis weight) (3) Update weight for each data (example distribution) • Weak Classifier: < 50% error over any distribution

  21. Adaboost: Design of weak classifier (1/2)

  22. Adaboost: Design of weak classifier (2/2) • Select a weak classifier with thesmallest weighted error • Prerequisite:

  23. Adaboost • Strong classifier = linear combination of T weak classifiers (1) Design of weak classifier (2) Weight for each classifier (Hypothesis weight) (3) Update weight for each data (example distribution) • Weak Classifier: < 50% error over any distribution

  24. Adaboost: Hypothesis weight (1/2) • How to set ?

  25. Adaboost: Hypothesis weight (2/2)

  26. Adaboost • Strong classifier = linear combination of T weak classifiers (1) Design of weak classifier (2) Weight for each classifier (Hypothesis weight) (3) Update weight for each data (example distribution) • Weak Classifier: < 50% error over any distribution

  27. Adaboost: Update example distribution (Reweighting) y * h(x) = 1 y * h(x) = -1

  28. Reweighting In this way, AdaBoost “focused on” the informative or “difficult” examples.

  29. Reweighting In this way, AdaBoost “focused on” the informative or “difficult” examples.

  30. Summary t = 1

  31. Example 2

  32. Example (1/5) Original Training set : Equal Weights to all training samples Taken from “A Tutorial on Boosting” by Yoav Freund and Rob Schapire

  33. Example (2/5) ROUND 1

  34. Example (3/5) ROUND 2

  35. Example (4/5) ROUND 3

  36. Example (5/5)

  37. Example 3

  38. Example 4

More Related