SIMPLE-MODELS FOR READING NEURONAL POPULATION CODES

In many neural systems, sensory information is distributed throughout a population of neurons. We study simple neural network models for ext racting this information. The inputs to the networks are the stochasti c responses of a population of sensory neurons tuned to directional st imuli. The performance of each network model in psychophysical tasks i s compared with that of the optimal maximum likelihood procedure. As a model of direction estimation in two dimensions, we consider a linear network that computes a population vector. Its performance depends on the width of the population tuning curves and is maximal for width, w hich increases with the level of background activity. Although for nar rowly tuned neurons the performance of the population vector is signif icantly inferior to that of maximum likelihood estimation, the differe nce between the two is small when the tuning is broad. For direction d iscrimination, we consider two models: a perceptron with fully adaptiv e weights and a network made by adding an adaptive second layer to the population vector network. We calculate the error rates of these netw orks after exhaustive training to a particular direction. By testing o n the full range of possible directions, the extent of transfer of tra ining to novel stimuli can be calculated. It is found that for thresho ld linear networks the transfer of perceptual learning is nonmonotonic . Although performance deteriorates away from the training stimulus, i t peaks again at an intermediate angle. This nonmonotonicity provides an important psychophysical test of these models.