Dr Mikko Juusola: Information theoretical methods

Entropy and Noise Entropy of Analogue Signals Shannon’s formula (1948) for calculating a neuron’s information capacity is valid only if the signal and noise components of the voltage responses are linear and Gaussian. It fails with non-linear, non-Gaussian responses of visual neurones under natural stimulation. Our new triple extrapolation method (Juusola & de Polavieja, 2003) is not restricted by these limitations. The information rate of a photoreceptor is calculated from the statistical properties of its voltage responses. The method provides a direct measure of how much information is contained in a neural response and is independent of any assumptions about what the neural responses represents. Thus, the information could be encoded at any temporal resolution and could involve any kind of temporal pattern. For example, with the method we are investigating:

(1) The effect of experience and rearing on the goodness of signalling at various natural stimulus conditions that are designed either to promote sensitivity (requiring spatio-temporal pooling of signals to resolve objects) or rapid adaptability (requiring effective redundancy reduction to prevent saturation). (2) If the reproducibility of Drosophila photoreceptor or LMC responses increases as the statistics of the visual input gradually approximate those of natural stimuli.