Fig Case Study

875 Words4 Pages
Referring to Fig. 1, mapping takes place from an input data space which may have more than 2 dimensions, onto typically a 2 dimensional array of neurons. Each neuron comprises a d dimensional weight vector (otherwise prototype vector or codebook vector) where the dimension of the input vectors is equal to d. Each neuron is connected to its adjacent neurons by a neighborhood relation, which determines the map structure or topology. The SOM can be thought of as a net which is spread to the data cloud. The SOM training algorithm adjusts the values of the weight vectors so that they span across the data cloud. A data item is then mapped into the node whose model is most similar to the data item, for example has the smallest distance from the…show more content…
This representation can help to visualize the clusters in the high-dimensional spaces, or to automatically recognize them using relatively simple image processing techniques.
The U-matrix will then be a 5x5 matrix with interpolated elements for each connection between two neurons as shown in Fig 4
The interpolated{x,y} elements represent the distance between neuron x and y, {4,5} equals the Euclidean distance between neurons (4) and(5)
SOM Configuration
A two-dimensional map was used as this allows the direct display of the subject’s condition before and after the Iron Man exercise. To simulate the SOM structure, the SOM Toolbox - CIS is used in the MATLAB environment. A hexagonal neighbourhood is used in this set of experiments, since this provides equidistant sets of neighbours and thus gives better results than the alternative square topology for visual displays. HRV data was recorded from athletes with two different physiological states: before and after they complete their Iron Man Triathlon. The input data set comprised the HRV measured over a 10 minute period, derived from the measured R-R times. When mapped the data formed clusters of similar HRV measurements. The map consists of 30 x 30 hexagonal topology neurons (yielding clearer results since each neuron has six immediate neighbours at same distance).
As may be seen there is a noticeable difference between the
Get Access