Neural Network Application Projects


Dr Chris Christodoulou

In the field of Artificial Neural Network applications, the areas that we are involved in are:

(i) Speaker identification for security systems in collaboration with researchers from King's and University Colleges, Univ of London and in liaison with an industrial company; 97% correct classification results have been obtained using different probabilistic RAM (pRAM) architectures (paper in IEEE Trans System Man & Cybernetics, 2001).

(ii) Face recognition using novel neural network techniques (in collaboration with researchers from Cyprus College - as a partner, funded by the Institute for Research Promotion, Cyprus). Related projects in this area that I am involved in are: age estimation from face images as well as extracting other text-based information from face images (e.g., gender, aspect ratio, colour, shape of chin) and face reconstruction from text-based description; reconstruction of missing facial features in partially obscured facial images.

(iii) Gas demand forecasting (with British Gas); our preliminary results with recurrent neural networks were very promising.

(iv) Time series prediction for financial modelling; currently a PhD student working on this subject (Mr C. M. Tang).

(v) Appraising London property prices; our preliminary results are encouraging.

(vi) Predicting football results; preliminary results for the English Premiership league give 57% prediction accuracy.

(vii) Drum sound recognition; 74% correct performance has been achieved.

(viii) Handwritten character recognition; 89.5% correct classification has been achieved with a combination of a Kohonen neural network and Principal Component Analysis (PCA).

(ix) Measuring the sugar Crystal Regularity Index (with Tate and Lyle Ltd); this index gives an indication of the crystal quality; preliminary results were encouraging and showed that the use of neural networks is feasible for this application.


Publications


Revised on 7th August 2002

webmaster@dcs.bbk.ac.uk