The Challenges of Visual Adaptation
Overview
The retina is tasked with the job of efficiently compressing visual information. Yet natural stimuli are composed of many challenges, one of which is the intensity range of the visual scene. With a limited range of response, adaptation is required to most efficient and reliably encode information. In this review paper, Dr. Rieke first discusses the aspects (statistical moments) of light distribution that the retina cares about, and focuses the rest of the paper on the competing requirements for adaptation mechanisms. Other interesting questions were also raised throughout the paper and are listed below.
Statistics of Neural Images
When encoding, naturally we would desire an optimal method to distribute the limited range of response to the input distribution so as to distinguish the difference between inputs, which Laughlin’s rule specifies. Simply speaking, the response should always be adjusted so that it can cover most of the input distribution while still being able to distinguish between similar stimuli. However, how do we adjust the stimulus-response curve in order to achieve such a goal? We would anticipate that changes in different statistical moments of the input would require different adjustments, and consequentially different mechanisms to achieve. Some argued that for fast adaptation, only the lower-order moments, i.e. mean and standard deviation, are of importance. To compensate for a shift in mean light intensity, the stimulus-response curve would have to be compressed in the x-axis (stimulus); and for changes in contrast, the slope would have to be modified. See figure 1-(c), (d) in the original paper linked below.
Competing Requirements for Adaptation
Speed is always a competing demand for any biological computation. However, the faster the operation, the lesser the accumulation of data, and thus the less accuracy. Therefore, spatial pooling is necessary for the reliability of inputs. It is unsurprising that at lower light levels, the effect of spatial pooling is more dominant due to low reliability of signals, hence the main adaptation occurs in the retinal circuitry. At higher light levels where the signal is more reliable, the adaptation switches to cones. Hence adaptation mechanisms switch during different conditions.
Other Interesting Questions
1. The location of dominant noise is important (is it before gain control or after?), and thus should be investigated carefully.
2. What is the nature of adaptation? Is it to hold noise constant? Or to hold the mean neural response constant?
3. Is adaptation really worth it when the signal is noisy enough? The fluctuations in signal may cause the resulting adaptation to be even worse than not adapting at all (see figure 7).
Summarized by: Pei Hsien Liu
The retina is tasked with the job of efficiently compressing visual information. Yet natural stimuli are composed of many challenges, one of which is the intensity range of the visual scene. With a limited range of response, adaptation is required to most efficient and reliably encode information. In this review paper, Dr. Rieke first discusses the aspects (statistical moments) of light distribution that the retina cares about, and focuses the rest of the paper on the competing requirements for adaptation mechanisms. Other interesting questions were also raised throughout the paper and are listed below.
Statistics of Neural Images
When encoding, naturally we would desire an optimal method to distribute the limited range of response to the input distribution so as to distinguish the difference between inputs, which Laughlin’s rule specifies. Simply speaking, the response should always be adjusted so that it can cover most of the input distribution while still being able to distinguish between similar stimuli. However, how do we adjust the stimulus-response curve in order to achieve such a goal? We would anticipate that changes in different statistical moments of the input would require different adjustments, and consequentially different mechanisms to achieve. Some argued that for fast adaptation, only the lower-order moments, i.e. mean and standard deviation, are of importance. To compensate for a shift in mean light intensity, the stimulus-response curve would have to be compressed in the x-axis (stimulus); and for changes in contrast, the slope would have to be modified. See figure 1-(c), (d) in the original paper linked below.
Competing Requirements for Adaptation
Speed is always a competing demand for any biological computation. However, the faster the operation, the lesser the accumulation of data, and thus the less accuracy. Therefore, spatial pooling is necessary for the reliability of inputs. It is unsurprising that at lower light levels, the effect of spatial pooling is more dominant due to low reliability of signals, hence the main adaptation occurs in the retinal circuitry. At higher light levels where the signal is more reliable, the adaptation switches to cones. Hence adaptation mechanisms switch during different conditions.
Other Interesting Questions
1. The location of dominant noise is important (is it before gain control or after?), and thus should be investigated carefully.
2. What is the nature of adaptation? Is it to hold noise constant? Or to hold the mean neural response constant?
3. Is adaptation really worth it when the signal is noisy enough? The fluctuations in signal may cause the resulting adaptation to be even worse than not adapting at all (see figure 7).
Summarized by: Pei Hsien Liu
Original Paper: Rieke, Fred, and Michael E. Rudd. “The Challenges Natural Images Pose for Visual Adaptation.” Neuron, vol. 64, no. 5, 2009, pp. 605–616., doi:10.1016/j.neuron.2009.11.028.
留言
張貼留言