Nandini D
5th BCA NCMS
SOM is developed by
Finnish professor and researcher Dr. TeuvoKohonen in 1982 was sometimes called
a Kohonen map. It is a type of artificial neural network which follows
unsupervised learning to generate a low dimensional (two dimensional),
discretized representation of the input space of the training samples are
referred as map, and is therefore a method to do dimensionality reduction. They
apply competitive learning instead as opposed to error correction learning.
Working:
The
data points in the data set recognize themselves by competing for themselves.
Initializing the weight vectors is the first step of mapping in SOM. A weight
that best represents that sample is searched from the randomly selected sample
vector and the map of the weight vectors. The weight vectors have neighboring
weights that are close to it. The chosen weight
and the neighboring weights are rewarded to become more like that
randomly selected sample vector. This will allow the SOM to take different
shapes. Commonly they form square/rectangle/hexagonal/L shape/in 2D feature
space.
Algorithm:
For 0
to X number of training epochs
Select a sample from the input set
Find the “winning” neuron for the
sample input
Adjust the weights of nearby neurons
End for
loop
Pros:
●
Data is interpreted and understood easily.
●
Dimensional reduction helps in observing
similarities in data.
Cons:
●
Extraneous data in weight vectors will add
randomness to the groups.
For more information about the SOM do watch Click the link below
No comments:
Post a Comment