In Part 1 of this series (DM Review, November 2000), I explained how neural networks operate and discussed their features and benefits. This month, I will discuss designing neural networks and their current and future trends.
The majority of neural network software runs under the Windows platform. More established vendors have made their software widely available on multiple platforms. For instance, MATLAB Neural Network Toolbox by The MathWorks Inc. runs on Apple Macintosh, Sun SPARC, DEC RISC, Dec ALPHA, SGI, HP9000 Series 300/400/700, IBM RS/6000 and VAX/VMS, CONVEX, CRAY and Pentium Series.
Some of the current issues of concern are scalability, testing, verification and integration of neural network systems into the modern environment. Many neural network applications become unstable when applied to larger problems. For example, government sectors such as defense, nuclear and space are concerned with testing and validation where accuracy is crucial. Most of the mathematical models and algorithms used to guarantee performance are still under development. Some experiments indicate that applied neural networks show poor performance when implemented on conventional computers, while others reveal instability to explain the results they obtain.
Even so, there have been many successful implementations of applied neural network software that innovative digital businesses have employed to solve a variety of problems. For example, a leading automotive manufacturing business adopted tools such as MATLAB, Simulink, Stateflow and Real-Time Workshop as a total design solution. Using these tools, the company has been able to produce a better product to market faster and cheaper. Additionally, they have a new pathway to innovation as evidenced in the company's release of a revolutionary hybrid electric vehicle.
In another example, a highly rated U.S. investment firm successfully manages more than $60 million dollars in investments by utilizing digital intelligence business models. They rely almost exclusively on computer techniques to guide their decisions in predicting the S&P 500 Index. The latest approach in forecasting used by the firm integrates an expert system with California Scientific Software's Brain-Maker neural network application. The expert system provides rules which govern the application of the neural network to the prediction. The neural network predicts the S&P 500 with an average accuracy of 95 percent. This statistic was obtained by testing the network on hundreds of days it had never seen before. The network is retrained every night with the most current information to keep its behavior in agreement with the current behavior of the market.
There is a broad range of existing and potential applications in the business environment where neural technology can be deployed. Neural networks are particularly suitable for tasks that would require a superior performance in classification, assessment, prediction and diagnosis. The following are a few examples:
Finance - Scorecard development and credit risk decision support, predicting bad debtors, preventing application fraud, predicting mortgage attrition, payment profiling, helping determine collection strategies, financial and economic forecasting, bond rating and market timing (MathWorks, Tradetrek, SIMUL8 Corp., Nestor, Neural Technologies, Cybersource, Metavante).
E-commerce - Online credit scoring in real time and online credit application fraud detection (eHNC).
Telecommunications - Trans-actional and application fraud prevention, forecasting potential bad debtors, customer profiling for loyalty/marketing, predicting churn, network performance monitoring, network capacity planning and simulation, channel equalization in mobile communications, network intrusion detection and anti-virus protection, and content filtering (Computer Associates, Symantec, Axeon).
Marketing - Predicting customer defection, profiling customers, behavioral scoring, analyzing consumer spending patterns and directing marketing strategy, general data mining, knowledge management (Autonomy, KMS, HNC Financial Solutions, SAP).
General Computing - Speech, text, pattern and optical character recognition (Neuroscript, Ncorp, Autonomy, Neuradynamics).
Industrial - Process control, power demand prediction, forensics, medical diagnosis, chemical formulation, materials, design and discovery, and music composition (KFx, Pegasus, Imagination Engines Inc.).
While corporate markets still have certain concerns about neural technology, more and more companies are investigating the possible deployment of neural networks for their business. For example, the governing body in the United Kingdom found that more than 70 percent of the largest companies were considering implementing such applications. Of those that had applied neural networks and other integrated digital competencies, 84 percent were satisfied with the results they achieved, finding savings in time and cost, and increased efficiencies.
Register or login for access to this item and much more
All Information Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access