1 / 12

Modular network SOM

Modular network SOM. Presenter : Cheng-Feng Weng Authors : Kazuhiro Tokunaga, Tetsuo Furukawa 2009/05/21. NN.9 (2009). Outline. Motivation Objective Method Experiments Conclusion Comments. Motivation. The conventional SOM can only deal with vectorized data.

lily
Download Presentation

Modular network SOM

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. Modular network SOM Presenter : Cheng-Feng Weng Authors :Kazuhiro Tokunaga, Tetsuo Furukawa 2009/05/21 NN.9 (2009)

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

  3. Motivation • The conventional SOM can only deal with vectorized data. • If one wishes to deal with a nonvector dataset, then one needs to make the data vectorized in advance or modify the SOM itself to adapt to the data type. Tokyo New York We only can know they are similar. But no more information about that.

  4. Objective • It develops a generalized framework of an SOM called a modular network SOM (mnSOM). • Every vector unit is replaced by a trainable functional module such as a neural network. Choose a module what you want

  5. The mnSOM with MLP modules • The MLP is multi-layer perceptrons. Distance measure…(1) Determine the BMM(BMU)…(2) Energy function for the SOM…(4) Learning weight…(3) Adapt the MLP using back-propagtion method…(5)

  6. The process of mnSOM Finished map (5)Update MLP Data 3 2 1 2 0.5 1 1 2 1 1 0.5 1 2 (1)(2)(3)(4) batch SOM

  7. Experiment 1 • The example if a family of cubic functions y=ax^3+bx^2+cx I=6, J=200 I=126,J=8

  8. Experiments 1(cont.)

  9. Experiment for weather map A period of 100days of the year2000 at 20 cities. The 10 cities for training and others for testing.

  10. Experiment for weather map(cont.) Preserving the topology of geo. map

  11. Conclusion • It’s also possible to vectorize the function shapes since the experimenter knows those functions in advance. • This means that the feature map of the cubic function family can be generated by the conventional SOM as well. • The advantages of mnSOMs: • Every module in an mnSOM has the capability of information processing. • The mnSOM also provides a way of fusing a supervised and an unsupervised learning algorithm.

  12. Comments • Advantage • Interesting expriments. • The concept is simple. • Drawback • Applying other modules is inconvenient. • Application • Time serial data.

More Related