Neural Networks and their Applications in Industry

INTRODUCTION

Over the past few years, technology has become very dynamic. It is fuelling itself at an ever-increasing rate. Computers are a prime component of this whole revolution. Computers that can help fight diseases by designing new drugs, computers that can design better computers, computers that simulate reality, and what not! This is a very exciting time for technology as the traditional boundaries are now becoming blurred. 

We often tend to think that computers can only decide on whether a statement is true or false. Such logical statements are then linked together to form a series of rules. To program a computer, all that is needed is to precisely define the problem, write a specification and use these rules. The program tells the computer rule by rule, exactly what to do. But it is difficult to program a computer for more subjective tasks, like predicting what the weather is going to be, or what the price of gold will be tomorrow. These tasks are in fact impossible to define accurately. Patterns need to be recognized that are complex and imperfect. Nature is chaotic and we need something to decode this chaos.

A different approach is needed to give computers more human-like abilities, the capability to make judgments, guesses and to change opinions. We humans learn by example and do not need to see every example to make a guess, a judgment based upon what we have been taught. 

The problem growing throughout the 1990s and into the millennium is that engineers no longer have the luxury in development to calculate all the algorithms or identify all the rules in these complex systems. In fact, most of these systems are so chaotic that doing so would be futile and prone to failure. 

Given the high stakes and intense competition within all areas of industry, intelligent business decisions are more important than ever. Even more important is the case for military applications. Data analysis plays an important role as a critical strategic weapon in the business and operations of the armed forces (both peace-time and war-time). The inherent limitations of existing statistical technology make normal data analysis very tedious and often costly process-requiring assumptions, rigid rules, force-fitting of data, as well as extensive trial and error experimentation and programming. Interpreted errors, biases, and mistakes are introduced. Valuable competitive insights are lost. Technology-based on artificial intelligence (Al) will soon become the only way to generate such systems economically. 

WHY ARTIFICIAL INTELLlGENCE/NEURAL COMPUTING?

The rapid pace of change and a climate of competitiveness, in which the profoundly and speedily informed gain the advantage and demand a more incisive consideration and induction of emerging IT than has hitherto been the case. There are a range of Al technologies available now-each with its own strengths and weaknesses. These are:

• Expert systems
• Fuzzy logic
• Case-based reasoning
• Neural networks
• Genetic algorithms 

NEURAL NETWORKS

Neural computers are based on the biological processes of the brain (human neural systems). Terms like can learn brain-like, massively parallel, learning machines and revolutionary have been used to describe neural computing. And these are true! It is not surprising that most industries believe that taking a neural approach will require special, expensive neural integrated circuits, big parallel computers, or very high-powered computers. This is not true! 

Conventional computers concentrate on emulating human thought processes, rather than actually how they are achieved by the human brain. Neural computers, however, take an alternative approach in that they directly model the biological structure of the human brain and the way it processes information (although at a much simpler level). This necessitates a new kind of architecture, which, like the human brain, consists of a large number of heavily interconnected processing elements operating in a parallel manner. Such an architecture is now both technically and commercially feasible to be deployed on a standard computer (from the laptop and desktop to the mainframe) and is certain to increase in general usage.

Neural networks are mathematical models, originally inspired by biological processes in the human brain. They are constructed from a number of simple processing elements interconnected by weighted pathways to form networks. Each element computes its output as a non-linear function of its weighted inputs. When combined into networks, these processing elements can implement arbitrarily complex non-linear functions which can be used to solve classification, prediction, or optimization problems

1.  Basic Theory - In this Section, neural networks are considered from an analytical viewpoint, so as to dispel any notions that neural networks are 'magical devices'. In fact, a neural network is little more than an example of a fairly specialized parallel processing architecture. A point that should be noted is that neural computing is not to be viewed as a competitor to conventional computing, but rather as a complementary technique. The most successful neural computing applications to date have been those which operate in conjunction with other computing techniques. For example, using a neural network to perform the first pass over a set of incoming data, then passing the results over to a conventional system for subsequent processing. 
2. What is a Neural Network? - Neural networks can be taught to perform complex tasks and do not require programming as conventional computers. They are massively parallel, extremely fast, and intrinsically fault-tolerant. They learn from experience, generalize from examples, and are able to extract essential characteristics from noisy data. They require significantly less development time and can respond to situations unspecified or not previously envisaged. They are ideally suited to real-world applications and can provide solutions to a hos' of currently impossible or commercially impractical problems. In simple terms, a neural network is made up of a number of processing elements called neurons, whose interconnections are called synapses. Each neuron accepts inputs from eitl~er the external world or from the outputs of other neurons. Output signals from all neurons eventually propagate their effect across the entire network to the final layer where the results can be output to the real world. The synapses have a processing value or weight, which is learned during the training of the network. The functionality and power of the network primarily depend on the number of neurons in the network, the interconnectivity patterns or topology, and the value of the weights assigned to each synapse.

APPLICATIONS OF NEURAL NETWORKS

Artificial neural networks have become an accepted information analysis technology in a variety of disciplines. This has resulted in a variety of commercial applications (in both products and services) of neural network technology (The applications that neural networks have been put to and the potential possibilities that exist in a variety of civil and military sectors are tremendous.) 

1. Forecasting the Behaviour of Complex Systems is a broad application domain for neural networks. Specific examples include electric load forecasting, economic forecasting, and forecasting natural and physical phenomena. One of the recent applications being studied is river-flow forecasting. It is an important application that can have a significant economic impact. It can help in predicting agricultural water supply and potential flood damage, estimating loads on bridges, etc. 
2. Signal Processing - Over the past decade or so, neural network approaches have been successfully combined with other signal processing techniques to produce a wide variety of applications. It can very well be argued that the commercial success of neural networks has been from its ready incorporation into other information processing approaches, such as pattern recognition and statistical inference, as well as symbolic processing. 
3. Data Compression - A class of neural networks called the backpropagation network (BPN) is useful in addressing diverse problems requiring recognition of complex patterns and performing non-trivial mapping functions. Data compression is a common problem in today's world. Specifically, one would like to find a way to reduce the data needed to encode and reproduce accurately a moderately high-resolution video image, so that these images may be transmitted over low-to-medium bandwidth communication equipment. Although there are many algorithmic approaches to perform data compression, most of these are designed to deal with static data, such as ASCII text, or with display images that are fairly consistent, such as computer graphics. Because video data rarely contains regular, well-defined forms (and even less frequently contains empty space), video data compression is a difficult problem from an algorithmic viewpoint. On the other hand, a neural network approach is ideal for a video data-reduction application, because BPN can be trained easily to map a set of patterns from an n-dimensional space to an m-dimensional space. Since any video image can be thought of as a matrix of picture elements (pixels), the image can be conceptualized as a vector in n-space. If we limit the video to be encoded to monochromatic, images can be represented as vectors of elements, each representing the grayscale value of a single pixel. 
4. Paint Quality Inspection - Visual inspection of painted surfaces, such as automobile body panels, is currently a very time-consuming and labor-intensive process. To reduce the amount of time required to perform this inspection, one of the major US automobile manufacturers reflects a laser beam off the painted panel and onto a projection .screen. Since the light source is a coherent beam, the amount of scatter observed in the reflected image of the laser provides an indication of the quality of the paint finish on the car. In the past, an inspection of scatter-pattern would have been performed primarily by humans, because conventional computer programming techniques that could be used to automate the observation and scoring process suffered from a lack of flexibility, and were not particularly robust. By using a backpropagation type of neural network to perform the quality-scoring operation, a system can be constructed that captures the expertise of human inspectors and is relatively easy to maintain and update. To improve the performance of the system, algorithmic techniques can be coupled with the above approach to simplify the problem. 
5. DNA Sequence Analysis - In the 45 years since the discovery of DNA's helical structure, scientists have made great strides in exploring human DNA's structure and locating human genes. As a part of this effort, advanced recombinant DNA and gene-mapping techniques, developed over the last two decades, have led to an unprecedented effort to map and sequence the entire human genome-the collection of 1,00,000 human genes. The huge amount of data the human genome project (presently going on in the US with scientists all over the world participating) produced will require high-performance computing and a more intelligent computer algorithm for analysis and inference. Recently, the neural network model has been recognized as a promising Al technique because such approaches might well embody important aspects of intelligence not captured by symbolic and statistical methods. These knowledge-based neural networks, called expert networks in some cases, perform as well as human experts (and often exhibit characteristics of a traditional symbolic expert system)

 CONCLUSION 

Neural computers perform very favorably in business and military applications. They do not require explicit programming by an expert and are robust to noisy, imprecise, or incomplete data. Furthermore, knowledge is encapsulated in a compact, efficient way that can easily be adapted to changes in a business environment.

As with all technologies, there is a window of opportunity for exploitation-and that window is here today. You cannot afford to ignore the fact that your competitors are already investigating the opportunities and realizing the significant business benefits that neural technology brings to a range of applications.

The reason one should use neural computing technology is the competition!