An adaptive analysis of neural encoding

Neural systems encode information about biological stimuli in the firing patterns of their ensembles. Mathematical algorithms designed to decode these firing patterns offer one approach to deciphering how neural systems represent and transmit this information. Current methods analyze spike train decoding as a two-stage process. The first, or encoding stage, characterizes the relation between the stimulus and the neural response. In the second, or decoding stage the stimulus is estimated from the neural firing patterns. Accuracy of the decoding analysis depends critically on how well the stimulus-response relation is characterized in the encoding stage. Stimulus encoding by a neural system is dynamic in that the information representation changes with changes in the stimulus and the synaptic connections among the neurons in the ensemble. None of the current decoding paradigms analyzes the dynamics of encoding. To study the dynamics of neural encoding we develop an adaptive parameter estimation algorithm. We use a Gaussian stimulus-Poisson response model of neural activity to develop the algorithm and perform a stability analysis. An extension of the adaptive paradigm to the analysis of spatial information encoding by place cells in the rat hippocampus is described