Artificial Neural Network
Presented By: Anirban RoyUnder the guidance of:Dr. Mrutyunjaya Panda
Gandhi Institute for Technological Advancement,
Bhubaneswar
Just as life attempts to understand itself better by
modeling it, and in the process create something new, so Neural computing is an attempt at modeling the
working of a brain and this presentation is an attempt to understand the basic concept of Artificial Neural Network.
Contents
• Brief Historical Background• What exactly is a Neural Network?• Structure of Artificial Neural Network (ANN)• Architecture & Design of ANN• Learning Processes in ANN
• Applications of ANN• Limits to Neural Networks• Future of ANN• Conclusion
Brief Historical Background
• McCulloch and Pitts introduced first neural networking computing model in 1940.
• Rosenblatt’s work resulted in two-layer network model in 1950.
• It was capable of learning certain classifications by adjusting connection weights.
• Demerits:– It was not capable of solving classical
XOR problem.
What exactly is a Neural Network?
• Powerful data modeling tool capable of capturing and representing complex I/O relationships.
• Composed of large no. of highly interconnected processing elements that are analogous to neurons.
• Most common Neural Network Model is the ‘Multilayer Percepton (MLP)’.
• The goal of this network is to create a model that correctly maps the input to the output using historical data so that the model can be used to produce the output when the desired output is unknown.
• Neural Network resemble Human Brain in these ways:– A neural Network acquires knowledge
through learning.– A ANN’s knowledge is stored within inter-
neuron connection strengths known as synaptic weights.
– ANN modify own topology just as neurons in the brain can die and new synaptic connections can grow.
Neural Network?
Structure of Artificial Neural Network (ANN)
• Described by Frank Rosenblatt’s theory in 1958.
• Basic element of ANN is Percepton.
• Percepton has 5 basic elements:– n – vector– Weights– Summing function– Threshold device– An output (+1/-1)
• The threshold has a predefined setting.• If ‘Summation < Threshold’ implies O/P= -
1• If ‘Summation > Threshold’ implies O/P=
+1
Architecture of ANN• Feed-forward Network:
– Uni-directional Signal Flow without Feedbacks
– O/P of one layer doesn’t affect other layers
– Used in pattern recognition– Referred to as bottom-up or
top-down organization.
• Feedback Networks:– Contains feedbacks & dynamic
in nature– Powerful and extremely
complicated Network– Referred to as interactive or
recurrent– State is always changing in
nature
Network Layers or Design of ANN
• Activity of I/P units represents raw information.
• Activity of hidden unit is determined by activities of I/P units and the weights on the connections between I/P and hidden units.
• Behavior of O/P unit depends on activity of hidden units and weights between hidden and O/P units.
• Hidden units are free to construct their own representations of the I/P.
Learning Processes in ANN
Applications of ANN
• Character Recognition• Image Compression• Stock Market• Food Processing• Medicine• Target Recognition• Machine Diagnostics• Signature Analysis• Monitoring• Airline Security Control
Airline Security Control Practical Example
• Airports use ANN to screen for plastic explosives.
• Luggage is bombarded with neutrons and gamma ray re-emitted are recorded and fed to an ANN.
• The received value is compared to the possible or approximated value for safe goods (since explosives are rich in nitrogenous compounds).
• Explosives are detected with 95% probability.• To minimize classification error, supervised
training was conducted for the ANN.• The entire security system can handle 600 to
700 bags per hour.• The ANN raises false alarms on only 2
percent of the harmless bags.
Limits to Neural Networks
• Neural Network programs sometimes become unstable when applied to larger problems.
• Mathematical theories used to guarantee the performance of ANN is still under development.
• Rapid increase in processing time requirements as size of the problem expands.
• Unable to explain any results that they obtain.
• Network function as “black boxes” whose rules of operation are completely unknown.
• The ANN needs to go through rigorous training and learning periods for more efficient results.
• Operational problem encountered when attempting to simulate the parallelism of neural networks.
Future of ANN• Robots that can see, feel and predict the world
around• Improved stock prediction• Common usage of self-driving cars• Composition of music• Handwritten documents to be automatically
transformed into word processing documents• Study trends found in Human genes.• Self-diagnosis of medical problems.• More intelligent computer systems and other
machines.
Conclusion