Paired Learners for Concept Drift - PowerPoint PPT Presentation

Loading...

PPT – Paired Learners for Concept Drift PowerPoint presentation | free to download - id: 1847ab-ZDc1Z



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Paired Learners for Concept Drift

Description:

Dynamic weighted majority: An ensemble method for drifting concepts. Accuracy Weighted Ensemble ... Mining concept drifting data streams using ensemble ... – PowerPoint PPT presentation

Number of Views:52
Avg rating:3.0/5.0
Slides: 15
Provided by: makingCsi
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Paired Learners for Concept Drift


1
Paired Learners for Concept Drift
  • Stephen H. Bach
  • and Marcus A. Maloof
  • ICDM 2008

2
Outline
  • Introduction
  • Algorithm Paired Learner
  • Related work
  • Experimental study
  • Analysis and discussion
  • Conclusion

3
Introduction
  • Concept may drift
  • Stable learner
  • predicts based on all of its experience.
  • Reactive learner
  • predicts based on its experience over a short,
    recent window of time

4
Algorithm Paired Learner
5
Algorithm Paired Learner (cont.)
6
Related work
  • Streaming Ensemble Algorithm
  • W. N. Street and Y. Kim.
  • A streaming ensemble algorithm (SEA) for
    large-scale classification.
  • Dynamic Weighted Majority
  • J. Z. Kolter and M. A. Maloof.
  • Dynamic weighted majority An ensemble method for
    drifting concepts.
  • Accuracy Weighted Ensemble
  • H. Wang, W. Fan, P. S. Yu, and J. Han.
  • Mining concept drifting data streams using
    ensemble classifiers.

7
Experimental study
  • Base Learner Naïve-Bayes (NB)
  • Instance
  • distributions for each class Ci
  • prior probabilities P(Ci)
  • conditional probabilities P( xj Ci)
  • most probable class, C

8
Experimental study
9
Experimental study
10
Experimental study
11
Experimental study
12
Analysis and discussion
  • Moment of drifts
  • If the paired learner replaces S near these
    times,
  • then it should classify subsequent examples more
    accurately.
  • f(n) training time, g(n) classifying time
  • The time to train a paired learner on n examples
    is
  • O(nf(w)n g(1)), where w is the size of Rws
    window.
  • If the base learner retracts examplesO(n f(1)
    n g(1)).

13
Analysis and discussion (cont.)
  • Economy
  • Performed comparably but with less learners than
    other algorithms
  • Robustness to noise
  • Equal to other algorithms

14
Conclusion
  • Paired Learner
  • We introduced the notion of a paired learner,
    which uses the difference in performance between
    a reactive learner and a stable learner to cope
    with concept drift.
  • The properties of reactive learners when they as
    indicators
About PowerShow.com