What is neural network ?

Introduction:

Neural Network or an Artificial Neural Network(ANN) is a computational model which is based on biological neural network’s structure and function. Structure of Artificial Neural Network(ANN) is affected by the information which flows through network because neural network changes based on the input and output.

What Is IOT ? – Click Here

What Is BlockChain – Click Here

Aim Of Neural Networks

The aim is to develop a system to perform numerous computational tasks faster than the traditional system. As artificial neural network achieves a large collection of units which are interconnected in some order to perform communication between the units. These units are termed as nodes or neurons. Every neuron is connected through a connection link with other neuron. Each connection link is associated with a weight and weight contains the information about input signal. Neurons solve a particular problem using this information because weight stimulates or retards the signal that is being communicated. Each neuron has activation signal which is its internal state. After combining input signals and activation rule, output signal is generated and may be sent to the other units.

What Is IOT ? – Click Here

What Is BlockChain – Click Here

History of Neural Network:

History of Neural Network is divided into three eras:

ANN during 1940-1960:

 In 1943:  Two mathematicians Warren and Walter Pitts invented a circuitry system focused to approximate the functioning of the human brain that ran simple algorithm.

1956: Taylor introduced an associative memory network.

1957: Cornell University researcher Frank Rosenblatt built the Perception to perform advanced pattern recognition.

1960: Two models “ADALINE” and ”MADALINE” was proposed by Bernard Widrow and Marcian Hoff.

ANN during 1960-1980:

1961: “Back propagation” scheme for multi layer network is developed by Rosenbalt but it was an unsuccessful attempt.

1969: MIT researchers Marvin Minsky and Seymour Papert published a book “Perceptrons” which highlights the various issues with neural networks.

1971: Associative memories are developed by Kohonen.

1976: Adaptive resonance theory was developed by Stephen Grossberg and Gail Carpenter.

ANN from 1980-Present:

1982: Hopfield developed Energy approach.

1985: Ackley, Hinton and Sejnowski were developed Boltzmann machine.

1986: Generalized Delta Rule was proposed by Rumelhart, Hinton and wiliams.

1988: Kosko proposed Binary Associative Memory(BAM) and also introduced the Fuzzy logic concept in ANN.

What Is IOT ? – Click Here

What Is BlockChain – Click Here

Artificial Neural Network Model:

The below diagram states the general model of ANN followed by its processing.

 

Net input can be calculated from above diagram as follows-

 

Yin = x1* w1   + x 2*w2 + x3*w3 +……………………………………+  xm*wm.

 

By applying activation function output can be calculated-

 

O = F(Yin)

 

Applications of Neural Network:

1. Neural Network in Medicine:

neural network in medicine

In Medical field Artificial Neural Networks are currently trending research area. Recently research is done on the disease recognition from various scans like cardiograms, CAT scans, ultrasonic scans etc.

2. Cardiovascular systems modelling and diagnosing:

Neural networks are implemented experimentally to model the human cardiovascular system. Creating a prototype of the cardiovascular system of an individual and then comparing it with the real time physiological measurements that is taken from the patient’s diagnosis can be attained. If this procedure carried out routinely  then the potential harmful medical conditions can be encountered at early stage which makes the process of combating the disease much easier.

2. Electronic Noses:

 Electronic noses have various applications in telemedicine. Telemedicine provides the medicine over long distance via a communication link. In the remote surgical environment the electronic nose can identify the odour. These identified odours then electronically transmitted to the other place where the door generation system will recreate again.

What Is IOT ? – Click Here

What Is BlockChain – Click Here

4. Instant Physicians:

An application developed in the 1980’s named as “instant physicians” trained an auto-associative memory neural network for storing  a large number of medical records, each of which contains information of symptoms, diagnosis and treatment for a particular case. Even after training the net can be presented the inputs containing a set of symptoms so it will detect the full stored pattern that shows the best diagnosis and treatment.

5. Neural Network in Business: 

neural network in business

Neural Networks are used for the business purpose which includes resource allocation and scheduling. It is also used for database mining.

6. Marketing:  

AMT (Airline Marketing Tactician) is a computer system built of several intelligent technologies including expert system.  Neural network is combined with the Airline Marketing Tactician and trained using back propagation to simplify the marketing control of seat allocations of airline. A continuous adaptive solution is required for application’s environment which changes continuously and rapidly.Credit

What Is IOT ? – Click Here

What Is BlockChain – Click Here

7. Evaluation:

Various neural networks were developed by HNC company which built by Robert Hecht-Nielsen. One of them is credit scoring system that raised the profitability up to 27% of the existing system. This system was trained with 5048 applications and out of which 2597 were verified.

8. Neural Network in Signature Verification: 

signature verification using neural network

 

Neural Network plays a vital role in authentication of signature.Signatures are generally used to authorize and authenticate a person in legal transaction. A non-vision based technique is used to verify the signature. First approach is to extract the geometrical feature set which denote signature. In second step train the neural network with the help of an efficient neural network algorithm with the extracted feature set.  The signatures are genuine or forged are classified under the verification stage by the neural network.

9. Neural Network to Recognize Human Face:

recognize the human face using neural network

 It is an ordinary job because of characterization of “non-face” images. If a neural network trained well then only it can be divided into two classes named as “images have faces” and “images don’t have faces”. Initially all the images which are taken as input must be pre processed. After that images dimensionality  is to be reduced and finally classify it using neural network training algorithm.

10. Neural Network to Recognize Speech:

what is neural network_techfbi

 Speech takes an important role in human-human interaction. Thus people wants speech interface with computer. At present humans requires a sophisticated language to interact with machines and which are so difficult to learn and use. This communication can be easier if a spoken language can be used which is possible for machine to understand. ANN plays an prominent role in this area.

ANN used for recognizing speech are-

  • Multi-layered Network.
  • Multi-layered Network with the recurrent connection.
  • Kohonen’s SOM (kohonen’s Self-organised Maps).

      Kohonen’s SOM is prominently used network for speech recognition because it takes the input in small segments of speech waveform. This create the same type of phonemes as output array and this process is known as feature extraction technique. After feature extraction, it will recognize the utterance using some acoustic models.

Learn More Linkedin