Extraction of information from the dynamical activities of neural networks

Interacting with our environment requires to process huge amounts of sensory data in short time. This incoming information is combined with internal states and results in actions. The fundamental mechanisms behind this information processing are still not understood. Even how information is stored in and transmitted with sequences of action potentials is still under heavy debate. This thesis presents three new contributions: A theoretical method of processing information spike by spike in a fast and efficient fashion. It was shown that it is sufficient to use Poissonian neurons for performing fast and efficient information processing. A new mechanism, produced through selective visual attention, was revealed that renders information about different visual stimuli, represented in the activity of neuronal populations, more distinct. A method for neuronal-prostheses capable of protecting estimators of intended actions against non-stationaries, for the cost of an extra error signal.