Tid: 17 december 2000 kl 1615-1700
Plats : Seminarierummet 3733, Institutionen för matematik, KTH, Lindstedts väg 25, plan 7. Karta!
Föredragshållare: Mattias Karlsson
Titel: Optimal Information Transmission in the Auditory System
Sammanfattning:
This work is an investigation of the information processing performed in the first part of the neural auditory system. The basic tool used is a general class of performance criteria from statistics (the f-divergences) that have close relations to various fundamental bounds in information theory. In contrast to the often used signal-to-noise ratios (SNR), the f-divergences have a well-defined meaning for almost any kind of stochastic dynamical system, in particular nonlinear ones for which SNR can be misleading. Using maximization of f-divergences as performance criteria, optimum system behaviour is defined and analysed with respect to parameter and structural variations. In the numerical simulations it is found that, under the given assumptions, it is possible to define an optimal first part of the processing chain, which makes it easier for the the brain to detect a weak incoming signal in noise. The results also show that this optimal first part can be designed in several ways, but common to all suitable models is an intimate connection with the spontaneous spiking rate from the first neuron in the chain. This can therefore serve as a partial explanation of the up to now not completely accounted for rate of activity in an unstimulated auditory nerve.