Databases Reference
In-Depth Information
reactions on an artificial nervous system using a WSN showed that the GN
approach is able to differentiate between internal stress patterns in the net-
work and patterns that result from external loading conditions in a structural
health monitoring (SHM) application. In addition, the data storage capacity
requirements of a GN are low. Therefore, GN is most suitable for a WSN de-
ployment. Baig et al. [34] proposed using a GN pattern recognition algorithm
to detect a distributed denial of service (DDoS) attack in a WSN. The GN
algorithm was able to detect DDoS attack patterns in a WSN by analyzing
the internal tra
c flow of the network. This implementation of a GN has been
tested on three different network topologies, and the results have shown that
it produces high recognition accuracies for all topologies.
The GN algorithm also offers an energy-e
cient mechanism for pattern
recognition. This follows the work of Baqer and Khan [62] on energy-e
cient
pattern recognition approaches for WSNs. In their work, event detection based
on the GN was demonstrated. By conducting the detection and analysis in
situ, i.e., at the sensor node level, the GN was shown to offer an energy-
e
cient mechanism for event detection in WSNs. This is in contrast to existing
approaches, which perform the analysis at the base station.
The ability of the Graph Neuron (GN) algorithm to provide a fast, e
cient
and scalable solution for pattern recognition makes it suitable for deploy-
ment in a number of different network environments ranging from resource-
constrained networks, such as WSNs, to large-scale networks, such as the Inter-
net and peer-to-peer (P2P) networks. Nevertheless, a GN implementation has
its own limitations, including a large number of required neurons in large-scale
and multi-dimensional patterns and inaccuracies introduced by a phenomenon
known as the intersection or crosstalk problem. Given that the structure of
a GN network can be abstracted in the form of memory structure or actual
processing nodes working together to form a GN network, the first limitation
is less significant. The intersection problem is an important limitation of the
GN algorithm. This problem is a result of GN's inability to obtain full pat-
tern information. The GN builds up pattern information using links between
adjacent neurons. Learning or adapting information by means of adjacency re-
lationships between neurons is known as the comparative-collaboration tech-
nique for one-shot learning.
3.2 One-Shot Learning Model
Graph Neuron implements a one-shot learning approach in its recognition
procedure. In this learning approach, learning occurs collaboratively between
nodes rather than independently by each processing node, as is implemented
