International Journal of Advance Research and Innovation

 
Fig: 2. Feedback Control System in NN 
The delta learning rule is also referred to as widow-
hoff rule, named due to the originator. The delta learning 
rule is valid only for continuous activation function and in 
the supervised training mode. The learning signal for this 
rule is called delta. The delta rule may be stated as, “The 
adjustment made to a synaptic weight of a neuron is 

---

Volume 2, Issue 3 (2014) 676-683 
ISSN 2347 - 3258
International Journal of Advance Research and Innovation 
678 
IJARI 
proportional to the product of the error signal and the input 
signal of the synapse”. The delta rule assumes that the error 
signal is directly measurable. The aim of the delta rule is to 
minimize the error over training patterns. Delta rule can be 
applied for single output unit and several output units. In 
this learning, the output neural network complete among 
themselves to become active. The basic idea behind this rule 
is that there are a set of neurons that are similar in all 
aspects excepts for some randomly distributed synaptic 
weights, and therefore respond differently to a given set of 
input patterns. However, a limit is imposed on the strength 
of the neurons. This rule has a mechanism that permit the 
neuron to complete of the right to respond to a given subset 
of inputs, such that only one output neuron, or only one 
neuron per group, is active at a time. The winner neuron 
during competition is called winner-takes-all neuron. 
2.1. Applications of Artificial Neural Network 
Neural Network in arts: The rapidly expanding field of 
neural networks provides a new approach to computer art 
applications. Rather than operate in the traditional style of 
pre-programmed rule-following systems, neural networks 
have the power to learn to produce specific types of outputs 
from specific inputs, based on examples they are taught. 
Thus, instead of having to specify how to create a certain 
artwork, the artist can instead teach a network examples of 
the desired output, and have the network generate new 
instances in that style. Several applications have been found 
in Music area, ranging from psychological models of human 
pitch, chord, and melody perception, to networks for 
algorithmic 
composition 
and 
performance 
control. 
Generally speaking, the applications here (and in other 
fields) can be divided into two classes: “input” and 
“output”. The input side includes networks for recognition 
and understanding of a provided stimulus, for instance 
speech recognition, or modeling how humans listen to and 
process a melody. Such applications are useful for 
communication from human to machine, and for artistic 
analysis, of a set of inputs. The output side includes the 
production of novel works, applications such as music 
composition or drawing generation. As signal processors, 
networks should find wide application in image 
enhancement, color adjusting or altering, edge and line 
modification, texture processing, etc. all based on learned 
mappings from input pictures to desired outputs. The area of 
neural network handwriting recognition is also being widely 
explored. A network could be trained to take an image of a 
face as input, and produce as output an image of that face 
now displaying any of several chosen emotions: smiling, 
frowning, staring, tongue stuck out, etc. Sequential 
networks could be used to produce the step-by-step 
movements of objects following some trajectory in space. 
Networks can also find interesting use in real-world kinetic 
applications, for controlling the motion of robots and 
vehicles from single arms to self-driving automobiles. Art 
forms including kinetic sculpture, machine dance, 
automated music conducting, etc. are fertile areas for neural 
network applications.