A neural network approach to spatial reconstruction in the CTF detector

Artificial neural networks may in some cases present a new important approach to information processing. We have investigated whether the accuracy offered by this technique is good enough to extract physical information from the signals coming from an unsegmented large volume liquid scintillator detector. ticular, we wanted to understand whether this method is well suited to be implemented in the Borexino detector for monitoring or for on-line event selection purposes. The results obtained on data from a smaller scale Borexino-like detector, implementing a neural network algorithm on a sequential scalar computer, have been compared to those of a standard best-fit procedure.