Current Areas Of Application Neural Networks example essay topic

The Artificial Neural Networks: Introduction Artificial Neural Networks ANNs) are computation paradigms, in simple terms -computer based software systems, which implement simplified models of their bio logical counter parts i.e. the biological neural networks. They can be created and trained to analyze data to recognize trends based on observation data, immense applications exist in the financial, manufacturing, marketing, telecommunications, biomedical, and other domains. This essay would not delve into the technicalities instead, will adopt a simplified treatment to the subject. What it aims to do is to highlight their current and prospective applications in the unconventional fields of business and finance covering more specifically the capital markets, that form the subject of interesting ongoing research in the field of soft computing. Why neural networks? Neural networks, with their extraordinary capability to derive meaning from imprecise or complicated data, can be used to extract patterns and identify trends that are too intricate to be noticed by either humans or other computer techniques.

A "trained neural network" can be considered an "expert" in the type of information it has been given to scrutinize. Given new situations of interest, this "expert" can then be used to provide projections and solutions to the "what if" questions. Characteristics of ANNs Some of their distinct characteristics are: Artificial Neurons (ANs) show Local processing Rich connection patterns between the ANs provide for parallel processing. Learning from experience enables knowledge acquisition Knowledge storage in distributed memory the synaptic AN connection Supervised and / or Unsupervised behaviour Due to these distinct characteristics their advantages include: Adaptive learning: the ability to gain knowledge of how to perform tasks based on the data given for training or through initial experience.

Self-Organisation: An ANN can create its own representation or organisation of the information it receives during learning time. Real Time Operation: the computations through ANN can be carried out in parallel, and special hardware devices are being designed and manufactured which can exploit this capability. Fault Tolerance via Redundant Information Coding: Partial damage of a network leads to the corresponding degradation of performance. However, some network capabilities may be retained even with major network damage.

Learning in ANNs: These distinct characteristics are achieved owing to the ability of the neural networks to "learn". This could be both - supervised and unsupervised. Figure: Supervised Neural Network Supervised learning / training involves a mechanism of providing the network with the desired output either by manually "grading" the network's performance or by providing the desired outputs with the inputs. The network then processes the inputs and compares the resulting outputs vis-'a-vis the desired outputs. The errors are then propagated back to the system, causing the system to adjust the weights that control the network. This process occurs repeatedly as the weights are continually tweaked.

The "training set", is the set of data, which enables the training, which is processed many times during training as the connection weights are ever refined. Unsupervised learning / training: The network here is provided with inputs but the desired outputs are not given. The decision regarding the features to be used to group the input data is with the system. Figure: Unsupervised Neural Network This is often referred to as self-organization or 'adaption'. This aspect of the learning behaviour is an area of extensive research.

This adaption of the network to the environment, holds the promise to the enabling of the robots (that we so often come across in science fictions) to continually learn on their own as they encounter new situations and new surroundings. The current areas of application Neural networks have broad applicability to real world business problems and have already been successfully applied in many industries. Since neural networks are best at identifying patterns or trends in data, they are well suited for prediction or forecasting needs including: Industrial process control Target marketing Customer research Data validation Sales forecasting Risk management More specifically, ANN are used in applications that can range from recovery of telecommunications from faulty software to recognition of speakers in communications; from diagnosis of hepatitis to the interpretation of multi meaning Japanese words; mine detection undersea; texture analysis; 3 D-object recognition to handwritten word and facial recognition. Neural Networks and the business world The interesting trend that is coming up is that neural networks are making big inroads into the financial worlds. The decisions that credit card companies, Banking and lending institutions deal with involve learning and statistical trends and hence are generally not clear-cut. A loan officer is able to make a decision through the filled forms that are a part of the loan approval process.

This data is now being used by neural networks, which are trained on the data from past decisions, to meet government requirements. This information is based on what input, or combination of inputs, weighed heaviest on the decision. Similar back-propagation networks are being used to establish credit risks and credit limits by credit card companies. They find their use in the field of direct marketing, where their application to databases is being carried out to gain higher ordering rates through telemarketing etc. Neural networks are being used in all of the financial markets - stock, bonds, international currency, and commodities. Their use has reportedly reaped in huge benefits in the international financial markets like that of Japan.

For government bond transactions the Dai ichi Kang yo Bank reportedly saw a rise in their hit rate from 60% to 75% through the use of NNs. Figure: ANN Functionality in Finance markets Similar success stories are being narrated by research institutes and stock - prediction systems as well. [I] The capital markets They hold huge potential in terms of applications of Soft Computing techniques and hence especially the use of ANNs. Soft computing Adapted from the physical sciences SC represents an area of computing. Within this realm, Artificial Intelligence techniques attempt to solve problems by applying physical laws and processes.

This style of computing is particularly tolerant of uncertainty and imprecision, making the approach appealing to those researching within 'noisy' domains, where the signal-to-noise ratio is quite low. The three key areas of Fuzzy Logic, Neural Networks, and Probabilistic Reasoning (which includes Genetic Algorithms, Chaos Theory, etc) are included and find applications in investment trading in particular. The arena of capital markets is one such field where there is an abundance of noisy data. Hence the traditional computing typically gives way to soft computing, as the severe conditions applied by traditional computing cannot be met. This is predominantly evident where different outcomes are apparently invoked through the same sets of input conditions, or there is an abundance of poor quality or missing data. Consider the Operations of supervised NNs in Capital Market applications: Here the already known inputs are tried to e matched by the system with a target such as stock prices or bond ratings. in the beginning of the process a random set of weights are assigned to the connections between each set of neurons (which represent their intensity).

This enables the calculation of intermediate value (hidden layer) and the output of the system. The process halts when an optimum output sufficiently close to the target is achieved. Otherwise the weights are adjusted and the process continues. If the information flow is from the input to output the network is said to be feedforward.

Back-propagation is said to be involved when the inadequacies in the output are fed back to the network so that the algorithm can be improved. Such a network is called feed-forward NN with backward propagation (FINN with BP). Figure: Back Propagation ANN Thus ANNs find their indispensable use in the various capital market activities in the following way: Market forecasting Activities like projections of stock market indices, such as for example the Standard and Poor's 500 stock index, Treasury bill rates, net asset value of mutual funds fall under the purview of Market forecasting. The role of SC with the help of NNs in this case is to use quantitative inputs like technical indices and qualitative factors like political effects to automate stock-market forecasting and trend analysis. Trading rules The timing of investment is an imperative issue in determining the returns that are expected. To give an example the investment of one Rupee in 1-month treasury bills from 1990 to 2000 would give a much lower return as compared to the investment made of the same amount for the same time span in S&P 500.

But what is interesting to note is that, if the investment would have been made by switching between the two options, which ever is doing better, the returns would have been distinctly over whelming. Thus, due to this immense relevance of timing the trading rules have evolved that tend to optimize the buy-sell decisions of investors. There is ongoing research in this field to establish the profitability of market timing. The level of accuracy required to forecast market trends is beyond the ability of managers to attain. This is possible when SC with support from NNs, helps in creating a security trading decision support system which in an ideal environment is fully automated and triggered by both quantitative and qualitative factors. Bond ratings The Bond ratings are subjective opinions on the ability to service debt and interest by economic entities like finance companies, industries etc. that are published by standard bond rating agencies.

Several attempts have been made to model these using methods like linear regression and multiple discriminant analysis. ANNs have been used to derive a formula for the same and the models so developed are being worked on for further optimisation. The ANN models have been developed and are still being worked on for similar applications like portfolio construction, option pricing etc and hence are turning into a powerful tool to analyze the trends in the capital markets. [II] Business strategising: ANNs can be used to make strategic decisions for a company to increase its profitability or / and gain cost effectiveness. Amongst the many managerial applications that they have, their use for production planning and controls (PPC) has a potential to grow significantly because an ANN can support more than one function of a PPC system. Ideally a PPC has functions that range from forecasting in the framework of master production scheduling to capacity adjustment, from requirement planning regarding the consumption driven material to lot-size determination to aid production and procurement; and short term intervention into running processes Auditing balances The most important application in the recent times that has found its being through the use of ANNs is the Prediction of patterns for the auditing monthly balances of companies like those into manufacturing etc For a smooth process of auditing and managerial control of assets, the transactions are required to be recorded in a way that it's easy to recognize their impact in the balance sheet and the profit and loss statements.

This provides the management with an accurate view of the company; through the confirmation of account validity and financial reporting. The application of this ANN method to the company's analytical review process would allow auditors to recognize useful, unanticipated relationships between accounts, as well as supervising unusual fluctuations. Not only would there be increased objectivity and efficiency due to automation there would be fewer reporting errors. In the manual process, the actual balance would have to be compared with its expected value and the auditor would have to determine which accounts required further testing. With the advancement of modern technology, an excess of the receipts and accounting records i.e. the materials that the auditors review, are being electronically generated. As the amount of information increases, there is a need to handle and protect such information from threat of misuse through fraud.

Because ANN's can work with missing or incomplete data they are capable of discovering patterns in data that one might not have even thought to look for. So a beneficial byproduct is gained through the use of ANN, which is a proactive alert of patterns that decision-makers might not have even thought of looking for. The whole procedure involves decision steps like comparing the net sales, that need to be audited if found lesser than predicted value. Amongst the other variables that can be predicted; change in inventory and personnel cost value signify the optimum use of materials.

The trend of costs in operations can be checked by the administrative cost predictions. The gross margin would give an idea regarding the money that would be left to cover indirect costs and profit and similarly operating profits can be set for prediction to help management gain a better control on production by budgeting funds and allocating resources. Limitations Some issues of concern that regarding NNs today are the scalability issue, the testing and verification of practical implementation of theories and their integration into the modern environment. Large problems sometimes cause NNs to lose stability, which could pose threat in areas such as defence, space and nuclear industries. Apart from implementation issues, the high degree of parallelism required can cause operational problems.

Gazing Into the crystal ball: The trends emerging predict a continued migration from tools to applications, and a movement towards "hybrid systems". These systems will encompass other types of processes, such as fuzzy logic, expert systems, and kinetic algorithms. Considerable work is being put in by manufactures to develop what would be known as "fuzzy neurons". - a convergence of Fuzzy logic and ANNs. The integration of fuzzy logic with neural networks is an area of tremendous interest. In life most of the data doesn't exactly fit into a particular categories. A person can be very smart or very fat or not so tall etc. such real-world variations are taken into account by Fuzzy logic.

Thus together with neural networks, such systems would be able to solve real problems where a large part of the problem would be "fuzzy". the fuzzy neurons being developed would not simply give an answer in terms of affirmative-negative, they would try to accommodate the fuzziness of real life and accordingly They provide answers that are "fuzzy " as well. Such systems (FNs) may be initialized to what an expert thinks are the rules and the weights for a given application giving rise to an expert systems that utilizes the strength of all three disciplines to provide a better system than either can provide themselves Such "Hybrid systems" are what the future beckons.