The New and Computationally Efficient MIL-SOM Algorithm: Potential Benefits for Visualization and Analysis of a Large-Scale High-Dimensional Clinically Acquired Geographic Data
Algorithm 1
Presents the pseudocode for the MIL-SOM (Mathematically Improved Learning) Algorithm.
The MIL-SOM algorithm for training a 2-dimensional map is defined as follows:
Let
X be the set of n training patterns
W be a grid of units where and are their coordinates on that grid
be the best clustering after iterations where is the distance between all
possible pairs of neural nodes and data points
alpha be the original learning rate, assuming values in (0,1) initialized to a
given initial learning rate
alpha1 alpha*a1 be the first improved learning rate
alpha2 alpha*a2 be the second improved learning rate
a1 be the first nonnegative parameter of alpha1 when set to zero it yields the original SOM update
a2 be the second nonnegative parameter of alpha2 when set to zero it also yields the original SOM update
diff is the differentiation for
int is the integral term for with intervals 0 to (1 to
).
radius be the radius of the neighborhood function H ,
initialized to a given initial radius
Repeat
for to
for all , calculate absolute distance
for up to number_iteration
Calculate the sum of the distances between all possible pairs of
neural nodes and data points
Select the unit that minimizes as the winning neuron
Iterate to minimize the quantization and topological errors and select
