1 / 13

SOM of SOMs

SOM of SOMs. Presenter : Cheng-Feng Weng Authors : Tetsuo Furukawa 2009/07/09. NN.16 (2009). Outline. Motivation Objective Method Experiments Conclusion Comments. Motivation. The SOM provides a map of data vectors, but not a map of class distributions. Class confusion.

jarvis
Download Presentation

SOM of SOMs

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. SOM of SOMs Presenter : Cheng-Feng Weng Authors :Tetsuo Furukawa 2009/07/09 NN.16 (2009)

  2. Outline • Motivation • Objective • Method • Experiments • Conclusion • Comments

  3. Motivation • The SOM provides a map of data vectors, but not a map of class distributions. Class confusion

  4. Motivation (cont.) • Manifoldcan be seen a class distributions. linear manifold viewpoint manifold

  5. Objective • The paper is to propose a method of mapping class called SOMs that can represent the relationships between distributions. The manifold gradually changes shape. 15 classes

  6. The SOMs • It is a hierarchical structure of a set of child SOMs and a single parent SOM. Manifold Class distributions Bottle up

  7. The SOMs algorithm • There are J child SOMs and a parent SOM map. • Children and parent maps have own parameters. • Randomize parent SOM map, and use least qe map to replace child’s. • Class maps are estimated for each class dataset. • The BMMs are regarded as data vectors for parent map. • Update child’s weights by overwriting its BMM.

  8. An Example for the SOMs

  9. Experiments

  10. Experiments (cont.)

  11. Application to autonomous mobile robot

  12. Conclusion • The essence of the algorithm is to generate a higher rank of data representation with class information as a clue, and the given datasets are modeled by fitting to a fiber bundle.

  13. Comments • Advantage • From a class point of view • Inversed construction • Drawback • … • Application • Class manifold • LDA + SOM vs. SOM + LVQ

More Related