Input-controlled neural dynamics and its statistics in the olfactory system pattern processing

Computational abilities of the biological olfactory systems are often superior to the ones offered by the engineering pattern recognition techniques. However, the computational principles of the olfactory neural dynamics are not quite understood. In particular, it is unclear how an applied input (odor) affects the dynamics of the olfactory circuits. To make a progress towards our understanding of this issue this dissertation studies computational abilities of the input controlled dynamics of the neural systems and their relation to the olfactory system.;Based on the macro-dynamics of the olfactory bulb (OB), the mechanism of input-controlled bifurcation in a model of chaotic neuron is proposed. This mechanism enables the neural system to map the input space to the different modes of its own dynamics.;The principle of the input-controlled dynamics is then employed in the olfactory encoding scheme based on the statistical properties of the spike trains. It is proposed how the encoding with input-controlled interspike interval distributions can be implemented in a network of biologically realistic spiking neurons.;Additionally, pattern recognition by the olfactory cortex dynamics is investigated. In the developed model the temporal structure of the olfactory bulb input controls the spatial dynamics of the cortical neural ensemble.