Neural Technologies Ltd has set up an office in Singapore. Founder and chief executive of the Petersfield, UK-based company Nick Ryman-Tubb says South East Asia, and Singapore in particular, is so anxious to gain competitive advantage, particularly in the financial markets, that it is snapping up new technologies such as neural computing. The UK on the other hand, he said, is tending to be somewhat sluggish, not wanting to invest in the technology until it is tried and tested. This in spite of the UK Department of Trade & Industry’s three-year neural network awareness campaign (CI No 2,720), in which Neural Technologies played a major part. Founded in 1989, the company says its philosophy is to provide practical applications using neural networking techniques. Singapore, with its strong, dynamic economy is particularly interested in Neural Technologies’ financial applications, which offer financial forecasting, risk analyis, dollar-yen rate predictions, fraud recognition and customer modeling. The company uses its Advanced Modular Adaptive Network, which it likens to the human brain with its various specialist areas, for example vision, movement and speech. It no longer deploys one large neural network, but rather a series of smaller networks which tackle different areas of a problem, for example predicting the price of bonds in 30, 60 or 90 days. For fraud recognition, the company has produced what it calls a watchdog. This is a neural network that sits in between front and back office systems and records and reports on any unusual transactions, or patterns of behavior, alerting the company to possible abuse of the system. The company says that since the Barings Bank debacle, this is in considerable demand. The networks are also used increasingly by financial institutions to analyze customer data, enabling them to target products to the right customers. One large bank in Singapore is apparently taking this one step further, and using the method to poach customers from its competitors. According to the company, the networks can learn to spot tax dodgers, credit card fraudsters and even auditing anomalies in large company accounts.
No doubt they won’t be popular with every-one. Ryman-Tubb says the neural networks are achieving high degrees of accuracy in relation to traditional statistical modeling techniques. For a large financial institution to implement a Neural Technologies network, say for financial forecasting, it would probably set it back about 250,000 British pounds, about $375,000 which includes the consultancy, support and initial setting up of the network, although smaller systems in the order of 50,000 pounds, about $75,000 are also feasible for some applications. Ryman-Tubb admits that it is a substantial investment, but says it is one that South East Asian companies are prepared to make, and he believes that they will see the benefits in increased efficiency and more accurate predictions. He is very excited about the possibilities for neural computing in South East Asia, which he says is prepared to make long-term investment in the technology. UK financial markets, he says, are so slow in adopting it themselves that they are now in danger of being eclipsed by their Asian rivals. To try and propagate neural networking technology and appeal to a wider audience, the company is now targeting products at specific vertical markets. Neural Technologies has launched Minotaur, a fraud detection system for the mobile telecommunications and financial services industries. Using neural computing to ‘learn’ the behavior patterns of mobile phone users, Minotaur builds a unique ‘fingerprint’ of a user. In this way, the system can detect any abnormal patterns of usage and alert the company to possible fraudulant use. Minotaur uses all available fields of data in a database to develop a model for learning user patterns. It can also take data from multiple databases in use within an organization. The system then continually learns from the data, and continues to update itself, becoming more accurate at spotting fraud as time goes on. Thresholds can be set for mobile phone usage, to generate warning messages or automically bar calls. Due to the sensitive use of the system, Neural Technologies says Minotaur customers are reluctant to be identified. However, the company says it has sold the system to a large mobile telephone company in South Africa, and is talking to three of the four major network providers in the UK. It is also targeting the product at the financial services sector. Neural Technologies says Minotaur is not an off-the-shelf package and needs to be adapted to each user to fit in with their existing systems. It says it can run on anything from a stand-alone personal computer, to being fully integrated on a mainframe. Neural Technologies does an initial systems audit for customers, produces a prototype system, and then installs the final product. It could not give any precise costs of the system, but said the initial study might cost from 20,000 pounds, about $30,000 to 50,000 pounds, about $75,000 including the prototype, and the final system would go from 50,000 pounds and upwards.
By Joanne Wallen
