N.M. Branston*, W. El-Deredy** and P.J.G. Lisboa

*Institute of Neurology, London **Unilever Research Port Sunlight Laboratories

The tremors and limb stiffening characteristic of Parkinson’s disease patients may be alleviated by surgical lesions to the Globus Pallidus area of the basal ganglia. This procedure requires the patient to be awake during surgery, in order to carry out limb movements as required. A project led by the Institute of Neurology is concerned with the characterisation of the firing patterns measured from single cells in the basal ganglia, using self-organised maps and Hidden Markov modelling.

Generative Topographic Map (GTM) of single cell recordings from the basal ganglia, taken in the periods immediately preceding and following conscious limb movements.

Detailed GTM locations for individual cells in GPie, showing that they appear to have distinct firing patterns.

Forecasting marginal benefit

P.J.G. Lisboa

Neural networks offer an alternative modelling strategy for multivariate forecasting in the large customer databases characteristic of Banks and Insurance companies. This application example is typical of several commercial models developed for high profile financial institutions, where careful modelling and visualisation of multi-dimensional customer profiles has defined niche markets, while neural network predictions have matched the empirical performance of grouped data to within one standard deviation, thus contributing a performance advantage over linear multivariate classifiers for identical attribute features and customer base.

Illustration of the predicted marginal benefit for a large customer database, displayed against two of the more important attributes.