Two fundamental questions about the brain are, "What does it compute?" and "How does it represent information?" The retina serves as a tractable model system in which we can begin to answer these questions because it is isolated from the rest of the brain and because we can record from many cells at once . We have identified two classes of novel retinal computations that cannot be explained by current models. Following a periodic sequence of flashes, retinal ganglion cells fire a burst of spikes after the time of the missing flash . This response is only one member of a set of sophisticated forms of temporal and spatio-temporal pattern recognition performed in the retina . We also studied the retinal response to motion discontinuities, and we discovered a synchronized firing event among retinal ganglion cells that signals motion reversal . Another set of experiments was aimed at the question of information representation. The brain must interpret the spike trains from a population of ganglion cells in order to identify objects. We presented different shapes to the retina, and we studied the population code by performing the task of the brain: reading out shape identity from the spike trains of a population of cells. The results showed that more than 100 cells are required to reach the low error rates that are common in vision, and a read-out mechanism must use information about the correlations between cells to achieve high accuracy.
Two fundamental questions about the brain are, "What does it compute?" and "How does it represent information?" The retina serves as a tractable model system in which we can begin to answer these questions because it is isolated from the rest of the brain and because we can record from many cells at once . We have identified two classes of novel retinal computations that cannot be explained by current models. Following a periodic sequence of flashes, retinal ganglion cells fire a burst of spikes after the time of the missing flash . This response is only one member of a set of sophisticated forms of temporal and spatio-temporal pattern recognition performed in the retina . We also studied the retinal response to motion discontinuities, and we discovered a synchronized firing event among retinal ganglion cells that signals motion reversal . Another set of experiments was aimed at the question of information representation. The brain must interpret the spike trains from a population of ganglion cells in order to identify objects. We presented different shapes to the retina, and we studied the population code by performing the task of the brain: reading out shape identity from the spike trains of a population of cells. The results showed that more than 100 cells are required to reach the low error rates that are common in vision, and a read-out mechanism must use information about the correlations between cells to achieve high accuracy.