Data compression methods with using neurolike GTM structures

Applying neural networks for data compression provides compression with very good generalization characteristics with management losses or even without any information losses. In the basis of applying neural networks for data compression - is its training without supervisor (self-training).