Databases Reference
In-Depth Information
FIGURE 3.1: A two-dimensional GN network for a binary pattern of size five
bits.
In this chapter, we will discuss the aspects of one-shot learning and its
applicability in DPR applications. This discussion will also include the systems
and network considerations for a DPR deployment using a one-shot learning
approach. In addition, this chapter will further describe the Graph Neuron
(GN) approach (see Section 2.5.1), a one-shot learning DPR scheme designed
for WSN implementations [2, 34].
3.1 One-Shot Learning Graph Neuron (GN) Scheme
A GN network is built using a composition of inter-connected processing
nodes, known as Graph Neurons (GNs), which follow the size and dimension
of a given pattern. In its simplest form, a GN network forms a two-dimensional
array of neurons. Each neuron is labeled with a value and position, i.e., a col-
umn and row position. Figure 3.1 shows a GN network with a two-dimensional
array formation.
A GN network receives an input and stores or processes the input according
to the instruction received. Creating a method capable of parallel, in-network
processing was an emphasis in the development of the GN method. In con-
trast, other recognition algorithms are most often implemented using CPU-
sequential processing. The parallel, in-network processing capability allows
GN to perform fast recognition regardless of the size of the input patterns.
Furthermore, by disseminating patterns into pattern elements and distribut-
ing them across the network, the storage capacity of this approach is high.
According to Nasution [52], the GN algorithm was developed based on the hy-
pothesis that changing the design emphasis from high-speed sequential CPU
processing to parallel network centric processing will result in a better asso-
ciative memory resource.
