The CNN algorithm based on the -means clustering algorithm
is proposed. The proposed CNN algorithm has a strong
connection with some of the conventional unsupervised learning
algorithms. In order to obtain lower energy, the CNN dynamically
allocates the synaptic weights to the clusters with high energy.
Even though the CNN algorithm is not new when we consider
the -means algorithm, the CNN provides us with a new
interpretation of the -means algorithm and the relationships
with some of the conventional algorithms. While applying the
CNN algorithm to several problems, the CNN successfully converges
converges
to suboptimal solutions. Any undesirable local minimum
problem was not observed in our CNN experiments performed
with many different data sets. The CNN algorithm is applied to
several problems such as simple 2-D data problems and image
compression problems. When compared with conventional clustering
algorithms such as Kohonen’s self-organizing map and
Kosko’s differential competitive learning on these problems, the
proposed CNN algorithm produces comparable results with less
computational effort and is free of the optimum parameter selection
problem