[Neuronal network with modifiable synapses: decoding of composite sensory messages under unsupervised and permanent learning].

Afferent sensory fibers carry usually messages which combine several primitive entities. A model of neural structure under unsupervised and permanent learning is proposed, which can detect the primary independent entities mixed in the afferent message. The model is a network of neurons mutually interconnected by means of inhibitory modifiable synapses with efficacies locally controlled by a specific conjunction law of pre- and post-synaptic activities.