Pattern classificaton using neural network learning Paper1 (freely available) and Paper2 (Springer) talk about application of fuzzy cognitive map (FCM) for pattern classification. I am having a tough time to understand how the equation of FCM is constructed from a set of features. My application is in cognitive vision and I have features which are (hue values, SURF features) of an object that is being identified. How do I apply FCM equation  for pattern classification ? In these papers they have genetic algorithm as the learning technique. I need to apply neural network for learning and then use mean square error between the actual FCM and the simulated FCM. I simply cannot understand how to transform the features into nodes and construct the edges.
An illustration with an example will be really helpful. Thank you
 A: From the linked paper (Paper #1)

Each chromosome is defined as a floating-point vector, whose length corresponds
  to the number of variable in a certain problem. Each element in a vector is
  called as a gene and each chromosome consists of $N(N-1)$ genes, which are
  floating point numbers in the range $[-1, 1]$. When computing the FCM
  classifier's adjacent matrix, each gene stands for a non-zero weight value
  in matrix.

So in other words, the graph is composed of chromosomes. Specifically, each chromosome is considered to be a node, with entries which describes the edge weight to other nodes, so you may form an adjacency matrix from this data. Note that these entries are nonzero, so we have a complete graph, complete digraph if the FCM matrix $A$ is symmetric.
As you'll read in the paper, the matrix is randomly generated. To find the best method for representing your specific features as nodes and how to determine the edge weighting, you should consult the existing literature in your field, this is merely an example. It may also be of interest to investigate how you are currently representing your data, you may already have a suitable structure hidden in your current representation. 
Are you working in a narrow domain? It is possible that a Character Recognition problem would entail an entirely different representation to a Face Recognition problem. Specifically considering your problem will aid in finding the best representation. Consider the objects themselves, and the possible relation between them.
