From my understanding of the terms:

Self means:

  • No supervision is required during training.

Organizing means:

  • To create a topographic ordered map by using unsupervised competitive learning and cooperative learning.
  • Competitive: All neurons compete with each other to become a winning neuron. Mechanism known as ‘Winner take all learning’.
  • Cooperative: Neuron dealing with closely related piece of information come closer to each other.

Map means:

  • Neurons in output space attempt to map their weights to conform to the given input data. OR

  • The nodes in a SOM network attempt to become like the inputs presented to them. Because of this mapping, also called Self Organization Feature Map.

K-means, Hierarchical, DBSCAN all meet the definition of Self as it appears for SOM. We can also say they are Maps as there is mapping of input vector to centers. My question is: can we also say whether or not they are Organizing as well?

In brief, is the below table correct?enter image description here

  • 1
    $\begingroup$ I see neither the self-organizing aspect ("rearranging" data) nor the map (2d grid) aspect in any of the other algorithms. But I don't think this name is to be read as three features A and B and C. I don't think that table makes sense. $\endgroup$ Sep 15 '17 at 1:46
  • $\begingroup$ I get the impression that this is a semantics question, with the complication that these aren't really standard/well-defined terms. It's trying to interpret meanings from the title of some existing method, then map (heh) other methods onto them. It might be more fruitful to skip the interpretation part, and just start with a well defined property of interest. Then, ask whether it applies to specific methods. $\endgroup$
    – user20160
    Sep 15 '17 at 3:53
  • $\begingroup$ I know the properties of SOM. But 'Self organizing' means "No supervision is required during training. SOMs learn on their own through unsupervised competitive learning". Excluding 'Map, are traditional clustering algorithms 'Self organizing'. $\endgroup$ Sep 15 '17 at 9:19
  • $\begingroup$ Kmeans just uses one of the data points to represent a cluster. Kohonen contrasts kmeans as "accurate ordering" with SOM's "self-ordering". The SOM rearranges its topology dynamically, hence it is self organizing. "Supervised SOM" means you pass the classes along with the inputs--it's still SOM even if supervised. $\endgroup$ Sep 20 '17 at 6:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.