| All visual information we perceive is collected by our retina. Within a single visual scene, however, the range of light intensities may be several orders of magnitude greater than the range of outputs the retina can produce. To efficiently transmit information about the visual world to the brain, the retina must continually adjust its mapping from inputs to outputs so as to match the range of outputs to the current range of stimulus intensities. This solution, which we call 'adaptation,' is seen across species and across sensory modalities. Adaptation is driven by changes in the statistical parameters (e.g. mean, variance, etc.) of the distribution of inputs. These parameters are rarely explicitly present in the stimulus, however. An adapting system must therefore estimate these statistical parameters. Inherent in any estimation process is a tradeoff between speed and accuracy of estimation. This thesis investigates the consequences of such a tradeoff in the dynamics of visual adaptation. Chapter 3 provides experimental evidence that the dynamics of adaptation are limited by the underlying estimation process. We find that the rate or timescale of adaptation is dependent on both the stimulus history and the discriminability of a change in the statistical parameters of the stimulus. Chapter 4 investigates an optimal solution to the speed vs. accuracy tradeoff. We find that the dynamics of adaptation in the mammalian visual system closely match the prediction of this optimal tradeoff. Together, the results of this thesis provide a more complete view of the computation underlying neural adaptation in a dynamic world. |
