MEASUREMENT OF INFORMATION TRANSMISSION IN SPIKE TRAINS:DIRECT METHODS ANDSTIMULUS RECONSTRUCTION.

Rob de Ruyter van Steveninck

NEC Research Institute, Princeton NJ

As animals move through the world they often have to make decisions in real time based on a continuous stream of incoming sensory data. One would like to characterize neural coding and information transmission in a way that is relevant for this real time mode of operation. There are, broadly speaking, two ways to do this: 1 Direct methods. Here the idea is to quantify the reproducibility of the neural response to repeated presentations of an ongoing continuously varying stimulus. With some mild restrictions these methods quantify essentially all the information that is carried by a neuron, and they do not rely on knowing how the stimulus is encoded in the spike train.2 Reconstruction methods. The aim here is to measure the information from the joint distribution of stimulus and response, where both stimulus and response are considered functions of time. These methods rely on a model of how the stimulus is encoded in the spike train, or equivalently, how the stimulus can be decoded, or reconstructed, from the spike train.Methods of type 1 provide an upper bound on the information that can be obtained under 2, as there can never be more information encoded about the stimulus than the what is carried by the spike train. On the other hand, if one can find a reconstruction that carries an amount of information close to what is found under 1, one can be confident that one understands the neural code in the conditions of the experiment. In this talk I will review these methods, and present data in which we make the comparison between the two.