2920-Article Text-2895-1-10-20180903.pdf

International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169 Volume: 2 Issue: 1 96 – 100 ___________________________________...

International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169 Volume: 2 Issue: 1 96 – 100 _______________________________________________________________________________________ Research Paper on Basic of Artificial Neural Network Ms. Sonali. B. Maind Department of Information Technology Datta Meghe Institute of Engineering, Technology & Research, Sawangi (M), Wardha [email protected] Ms. Priyanka Wankar Department of Computer Science and Engineering Datta Meghe Institute of Engineering, Technology & Research, Sawangi (M), Wardha [email protected] Abstract—An Artificial Neural Network (ANN) is an information processing paradigm that is inspired by the way biological nervous systems, such as the brain, process information. The key element of this paradigm is the novel structure of the information processing system. It is composed of a large number of highly interconnected processing elements (neurons) working in unison to solve specific problems. ANNs, like people, learn by example. An ANN is configured for a specific application, such as pattern recognition or data classification, through a learning process. Learning in biological systems involves adjustments to the synaptic connections that exist between the neurons. This is true of ANNs as well. This paper gives overview of Artificial Neural Network, working & training of ANN. It also explain the application and advantages of ANN. Keywords:- ANN(Artificial Neural Network), Neurons, pattern recognition. _______________________________________________________*****_____________________________________________________ INTRODUCTION 4. Fault Tolerance via Redundant Information Coding: Partial destruction of a network leads to the The study of the human brain is thousands of years old. corresponding degradation of performance. However, With the advent of modern electronics, it was only natural to some network capabilities may be retained even with try to harness this thinking process. The first step toward major network damage. artificial neural networks came in 1943 when Warren McCulloch, a neurophysiologist, and a young Neural networks take a different approach to problem solving than that of conventional computers. Conventional mathematician, Walter Pitts, wrote a paper on how neurons computers use an algorithmic approach i.e. the computer might work. They modeled a simple neural network with follows a set of instructions in order to solve a problem. electrical circuits. Neural networks, with their remarkable Unless the specific steps that the computer needs to follow ability to derive meaning from complicated or imprecise are known the computer cannot solve the problem. That data, can be used to extract patterns and detect trends that restricts the problem solving capability of conventional are too complex to be noticed by either humans or other computers to problems that we already understand and know how to solve. But computers would be so much more useful computer techniques. A trained neural network can be if they could do things that we don't exactly know how to thought of as an "expert" in the category of information it do. Neural networks process information in a similar way has been given to analyse. Other advantages include: the human brain does. The network is composed of a large number of highly interconnected processing elements 1. Adaptive learning: An ability to learn how to do tasks (neurons) working in parallel to solve a specific problem. based on the data given for training or initial Neural networks learn by example. They cannot be experience. programmed to perform a specific task. The examples must 2. Self-Organisation: An ANN can create its own be selected carefully otherwise useful time is wasted or even organisation or representation of the information it worse the network might be functioning incorrectly. The receives during learning time. disadvantage is that because the network finds out how to 3. Real Time Operation: ANN computations may be solve the problem by itself, its operation can be carried out in parallel, and special hardware devices are unpredictable. On the other hand, conventional computers being designed and manufactured which take advantage use a cognitive approach to problem solving; the way the of this capability. problem is to solved must be known and stated in small unambiguous instructions. These instructions are then 96 IJRITCC | January 2014, Available @ http://www.ijritcc.org ______________________________________________________________________________________ International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169 Volume: 2 Issue: 1 96 – 100 _______________________________________________________________________________________ converted to a high level language program and then into machine code that the computer can understand. These machines are totally predictable; if anything goes wrong is due to a software or hardware fault. Neural networks and conventional algorithmic computers are not in competition but complement each other. There are tasks are more suited to an algorithmic approach like arithmetic operations and tasks that are more suited to neural networks. Even more, a large number of tasks, require systems that use a combination of the two approaches (normally a conventional computer is used to supervise the neural network) in order to perform at maximum efficiency. What is Artificial Neural Network? Traditionally neural network was used to refer as network Artificial Neural Networks are relatively crude electronic or circuit of biological neurones, but modern usage of the models based on the neural structure of the brain. The brain term often refers to ANN. ANN is mathematical model or basically learns from experience. It is natural proof that computational model, an information processing paradigm some problems that are beyond the scope of current computers are indeed solvable by small energy efficient i.e. inspired by the way biological nervous system, such as packages. This brain modeling also promises a less technical brain information system. ANN is made up of way to develop machine solutions. This new approach to interconnecting artificial neurones which are programmed computing also provides a more graceful degradation during like to mimic the properties of m biological neurons. These system overload than its more traditional counterparts. neurons working in unison to solve specific problems. ANN These biologically inspired methods of computing are is configured for solving artificial intelligence problems thought to be the next major advancement in the computing industry. Even simple animal brains are capable of functions without creating a model of real biological system. ANN is that are currently impossible for computers. Computers do used for speech recognition, image analysis, adaptive rote things well, like keeping ledgers or performing complex control etc. These applications are done through a learning math. But computers have trouble recognizing even simple process, like learning in biological system, which involves patterns much less generalizing those patterns of the past the adjustment between neurones through synaptic into actions of the future. Now, advances in biological research promise an initial understanding of the natural connection. Same happen in the ANN. thinking mechanism. This research shows that brains store information as patterns. Some of these patterns are very Working of ANN: complicated and allow us the ability to recognize individual The other parts of the ―art‖ of using neural networks revolve faces from many different angles. This process of storing around the myriad of ways these individual neurons can be information as patterns, utilizing those patterns, and then clustered together. This clustering occurs in the human mind solving problems encompasses a new field in computing. in such a way that information can be processed in a This field, as mentioned before, does not utilize traditional dynamic, interactive, and self-organizing way. Biologically, programming but involves the creation of massively parallel neural networks are constructed in a three-dimensional networks and the training of those networks to solve specific world from microscopic components. These neurons seem problems. This field also utilizes words very different from capable of nearly unrestricted interconnections. That is not traditional computing, words like behave, react, self- true of any proposed, or existing, man-made network. organize, learn, generalize, and forget. Integrated circuits, using current technology, are two- Whenever we talk about a neural network, we dimensional devices with a limited number of layers for should more popularly say ―Artificial Neural Network interconnection. This physical reality restrains the types, and (ANN)‖, ANN are computers whose architecture is scope, of artificial neural networks that can be implemented modelled after the brain. They typically consist of hundreds in silicon. of simple processing units which are wired together in a complex communication network. Each unit or node is a Currently, neural networks are the simple clustering of the primitive artificial neurons. This clustering occurs by simplified model of real neuron which sends off a new creating layers which are then connected to one another. signal or fires if it receives a sufficiently strong Input signal How these layers connect is the other part of the "art" of from the other nodes to which it is connected. engineering networks to resolve real world problems. 97 IJRITCC | January 2014, Available @ http://www.ijritcc.org ______________________________________________________________________________________ International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169 Volume: 2 Issue: 1 96 – 100 _______________________________________________________________________________________ These lines of communication from one neuron to another are important aspects of neural networks. They are the glue to the system. They are the connections which provide a variable strength to an input. There are two types of these connections. One causes the summing mechanism of the next neuron to add while the other causes it to subtract. In more human terms one excites while the other inhibits. Some networks want a neuron to inhibit the other neurons in the same layer. This is called lateral inhibition. The most common use of this is in the output layer. For example in text recognition if the probability of a character being a "P" is.85 and the probability of the character being an "F" is.65, the network wants to choose the highest probability and inhibit all the others. It can do that with lateral inhibition. Figure 1:- A Simple Neural Network Diagram. This concept is also called competition. Basically, all artificial neural networks have a similar Another type of connection is feedback. This is where the structure or topology as shown in Figure1. In that structure output of one layer routes back to a previous layer. An some of the neurons interfaces to the real world to receive example of this is shown in Figure 2. its inputs. Other neurons provide the real world with the network's outputs. This output might be the particular character that the network thinks that it has scanned or the particular image it thinks is being viewed. All the rest of the neurons are hidden from view. But a neural network is more than a bunch of neurons. Some early researchers tried to simply connect neurons in a random manner, without much success. Now, it is known that even the brains of snails are structured devices. One of the easiest ways to design a structure is to create layers of elements. It is the grouping of these neurons into layers, the connections between these layers, and the summation and transfer functions that comprises a functioning neural network. The general terms used to describe these characteristics are common to all networks. Figure 2:- Simple Network with Feedback and Competition. Although there are useful networks which contain only one layer, or even one element, most applications require networks that contain at least the three normal types of The way that the neurons are connected to each other has a layers - input, hidden, and output. The layer of input neurons significant impacton the operation of the network. In the receive the data either from input files or directly from larger, more professional softwaredevelopment packages the electronic sensors in real-time applications. The output layer user is allowed to add, delete, and control theseconnections sends information directly to the outside world, to a at will. By "tweaking" parameters these connections can be secondary computer process, or to other devices such as a made toeither excite or inhibit. mechanical control system. Between these two layers can be many hidden layers. These internal layers contain many of Training an Artificial Neural Network the neurons in various interconnected structures. The inputs and outputs of each of these hidden neurons simply go to Once a network has been structured for a particular other neurons. application, that network is ready to be trained. To start this process the initial weights are chosen randomly. Then, the In most networks each neuron in a hidden layer receives the training, or learning, begins. There are two approaches to signals from all of the neurons in a layer above it, typically training - supervised and unsupervised. Supervised training an input layer. After a neuron performs its function it passes involves a mechanism of providing the network with the its output to all of the neurons in the layer below it, desired output either by manually "grading" the network's providing a feedforward path to the output. (Note: in section performance or by providing the desired outputs with the 5 the drawings are reversed, inputs come into the bottom inputs. Unsupervised training is where the network has to and outputs come out the top.) make sense of the inputs without outside help. The vast bulk of networks utilize supervised training. Unsupervised training is used to perform some initial characterization on 98 IJRITCC | January 2014, Available @ http://www.ijritcc.org ______________________________________________________________________________________ International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169 Volume: 2 Issue: 1 96 – 100 _______________________________________________________________________________________ inputs. However, in the full blown sense of being truly self don't lock themselves in but continue to learn while in learning, it is still just a shining promise that is not fully production use. understood, does not completely work, and thus is relegated to the lab. 2. Unsupervised, or Adaptive Training. 1. Supervised Training. The other type of training is called unsupervised training. In unsupervised training, the network In supervised training, both the inputs and the is provided with inputs but not with desired outputs. The outputs are provided. The network then processes the inputs system itself must then decide what features it will use to and compares its resulting outputs against the desired group the input data. This is often referred to as self- outputs. Errors are then propagated back through the system, organization or adaption. At the present time, unsupervised causing the system to adjust the weights which control the learning is not well understood. This adaption to the network. This process occurs over and over as the weights environment is the promise which would enable science are continually tweaked. The set of data which enables the fiction types of robots to continually learn on their own as training is called the "training set." During the training of a they encounter new situations and new environments. Life is network the same set of data is processed many times as the filled with situations where exact training sets do not exist. connection weights are ever refined. The current Some of these situations involve military action where new commercial network development packages provide tools to combat techniques and new weapons might be encountered. monitor how well an artificial neural network is converging Because of this unexpected aspect to life and the human on the ability to predict the right answer. These tools allow desire to be prepared, there continues to be research into, the training process to go on for days, stopping only when and hope for, this field. Yet, at the present time, the vast the system reaches some statistically desired point, or bulk of neural network work is in systems with supervised accuracy. However, some networks never learn. This could learning. Supervised learning is achieving results. be because the input data does not contain the specific information from which the desired output is derived. Application Networks also don't converge if there is not enough data to enable complete learning. Ideally, there should be enough The various real time application of Artificial Neural data so that part of the data can be held back as a test. Many Network are as follows: layered networks with multiple nodes are capable of memorizing data. To monitor the network to determine if 1. Function approximation, or regression analysis, the system is simply memorizing its data in some non including time series prediction and modelling. significant way, supervised training needs to hold back a set 2. Call control- answer an incoming call (speaker-ON) of data to be used to test the system after it has undergone its with a wave of the hand while driving. training. 3. Classification, including pattern and sequence If a network simply can't solve the problem, the designer then has to review the input and outputs, the recognition, novelty detection and sequential decision number of layers, the number of elements per layer, the making. connections between the layers, the summation, transfer, 4. Skip tracks or control volume on your media player and training functions, and even the initial weights using simple hand motions- lean back, and with no need themselves. Those changes required to create a successful to shift to the device- control what you watch/ listen to. network constitute a process wherein the "art" of neural 5. Data processing, including filtering, clustering, blind networking occurs. Another part of the designer's creativity signal separation and compression. governs the rules of training. There are many laws 6. Scroll Web Pages, or within an eBook with simple left (algorithms) used to implement the adaptive feedback and right hand gestures, this is ideal when touching the required to adjust the weights during training. The most device is a barrier such as wet hands are wet, with common technique is backward-error propagation, more gloves, dirty etc. commonly known as back-propagation. These various 7. Application areas of ANNs include system learning techniques are explored in greater depth later in this identification and control (vehicle control, process report. control), game-playing and decision making Yet, training is not just a technique. It involves a (backgammon, chess, racing), pattern recognition (radar "feel," and conscious analysis, to insure that the network is not over trained. Initially, an artificial neural network systems, face identification, object recognition, etc.), configures itself with the general statistical trends of the sequence recognition (gesture, speech, handwritten text data. Later, it continues to "learn" about other aspects of the recognition), medical diagnosis, financial data which may be spurious from a general viewpoint. applications, data mining (or knowledge discovery in When finally the system has been correctly trained, and no databases, "KDD"). further learning is needed, the weights can, if desired, be 8. Another interesting use case is when using the "frozen." In some systems this finalized network is then Smartphone as a media hub, a user can dock the device turned into hardware so that it can be fast. Other systems to the TV and watch content from the device- while 99 IJRITCC | January 2014, Available @ http://www.ijritcc.org ______________________________________________________________________________________ International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169 Volume: 2 Issue: 1 96 – 100 _______________________________________________________________________________________ controlling the content in a touch-free manner from CONCLUSION afar. 9. If your hands are dirty or a person hates smudges, In this paper we discussed about the Artificial neural touch-free controls are a benefit network, working of ANN. Also training phases of an ANN. There are various advantages of ANN over conventional Advantages approaches. Depending on the nature of the application and 1. Adaptive learning: An ability to learn how to do tasks the strength of the internal data patterns you can generally based on the data given for training or initial expect a network to train quite well. This applies to experience. problems where the relationships may be quite dynamic or 2. Self-Organisation: An ANN can create its own non-linear. ANNs provide an analytical alternative to organisation or representation of the information it conventional techniques which are often limited by strict receives during learning time. assumptions of normality, linearity, variable independence 3. Real Time Operation: ANN computations may be carried out in parallel, and special hardware devices etc. Because an ANN can capture many kinds of are being designed and manufactured which take relationships it allows the user to quickly and relatively advantage of this capability. easily model phenomena which otherwise may have been 4. Pattern recognition is a powerful technique for very difficult or imposible to explain otherwise. Today, harnessing the information in the data and neural networks discussions are occurring everywhere. Their generalizing about it. Neural nets learn to recognize promise seems very bright as nature itself is the proof that the patterns which exist in the data set. this kind of thing works. Yet, its future, indeed the very key 5. The system is developed through learning rather than to the whole technology, lies in hardware development. programming.. Neural nets teach themselves the Currently most neural network development is simply patterns in the data freeing the analyst for more proving that the principal works. interesting work. 6. Neural networks are flexible in a changing REFERENCES environment. Although neural networks may take some time to learn a sudden drastic change they are Bradshaw, J.A., Carden, K.J., Riordan, D., 1991. excellent at adapting to constantly changing Ecological ―Applications Using a Novel Expert System information. Shell‖. Comp. Appl. Biosci. 7, 79–83. 7. Neural networks can build informative models Lippmann, R.P., 1987. An introduction to computing whenever conventional approaches fail. Because with neural nets. IEEE Accost. Speech Signal Process. neural networks can handle very complex interactions Mag., April: 4-22. they can easily model data which is too difficult to model with traditional approaches such as inferential N. Murata, S. Yoshizawa, and S. Amari, ―Learning statistics or programming logic. curves, model selection and complexity of neural 8. Performance of neural networks is at least as good as networks,‖ in Advances in Neural Information classical statistical modelling, and better on most Processing Systems 5, S. Jose Hanson, J. D. Cowan, problems. The neural networks build models that are and C. Lee Giles, ed. San Mateo, CA: Morgan more reflective of the structure of the data in Kaufmann, 1993, pp. 607-614 significantly less time. 100 IJRITCC | January 2014, Available @ http://www.ijritcc.org ______________________________________________________________________________________