A Kohonen self-organising network will be adopted as the AI
architecture. The self-organising map (invented by Teuvo
Kohonen) uses a form of unsupervised learning where a set of
artificial neurons learn to map points in an input space to
coordinates in the output space. The input space can have
different dimensions and topologies from the output space, and
the SOM (self-organising map) will attempt to preserve
these.
The idea behind such a system is to allow the intricacies of
the AI implementation to be hidden form the user, providing an
efficient way to catalogue data within the system.
Unsupervised learning is a method of machine-learning where a
model is fit to observations. It is distinguished from
supervised learning by the fact that there is no priori output.
In unsupervised learning, a data set of input objects is
gathered and input objects are typically treats as a set of
random variables.
In this case, the input objects will be a collection of
keywords to check the Euclidean space. A joint density model is
then built for the data set.
Unsupervised learning can be used to produce conditional
probabilities (i.e. supervised learning) for any of the
variables generated while querying the root complex database
table. Typically, this is classed under a feed-forward
architecture.
The algorythm:
1.Initialise the database and root complex table and cast it
as the SOM class.
2.Grab an input vector (i.e a keyword)
3.Traverse each node in the SOM
Use Euclidean distance formula to find similarity between
the input vector
and the map's node's
weight vector . Track the node that produces the smallest
distance (this node will be called the Best Matching Unit)
4.Update the nodes in the neighbourhood of BMU by pulling
them closer to the input vector Wv(t + 1) = Wv(t) +
˜(t)±(t)(D(t) - Wv(t))
5.During the training process, a map is built. The neural
network organises itself by pulling back data from the
database. The network must be given a large number of inputs or
keywords, so the ANN can retrieve several possible results from
the database. Otherwise, all input vectors must be administered
several times.
6.During the mapping process, a new input vector may quickly
be given a location on the map and will automatically be
classified or categorised. There will be one single winning
neuron, whose weight vector lies closest to the input vector
(this can be simply determined by calculating the Euclidean
distance between input vector and weight vector). Within the
proposed software, the root complex will be used as the
Euclidean space where all the meta-data will be stored. The
data will then be retrieved from the system in binary objects
which will be cast back to the specific class type. This class
will include encapsulated data. The closest match of relevancy
and accuracy will be presented to the user as the answer to the
question.
Printer friendly PDF
