On the interpretation of sensitivity analyses of neural responses

by Romain Brette
Abstract:
Responses of auditory neurons vary with many dimensions of acoustical stimuli. As a consequence, there is a difference between sensitivity to a particular dimension (e.g., ITD or level), which is assessed when only that dimension is varied while other dimensions are fixed (yielding tuning curves), and information about that dimension, which requires that all natural variability be considered. In particular, the rate of a neuron can be very sensitive to a dimension while poorly informative about it, if it is also sensitive to other dimensions. One implication is that in a multi-dimensional world, stimulus properties such as ITD are optimally coded with heterogeneous neural populations.
Reference:
Romain Brette, 2010.On the interpretation of sensitivity analyses of neural responses, The Journal of the Acoustical Society of America, volume 128.
Bibtex Entry:
@article{Brette2010,
 abstract = {Responses of auditory neurons vary with many dimensions of acoustical stimuli. As a consequence, there is a difference between sensitivity to a particular dimension (e.g., ITD or level), which is assessed when only that dimension is varied while other dimensions are fixed (yielding tuning curves), and information about that dimension, which requires that all natural variability be considered. In particular, the rate of a neuron can be very sensitive to a dimension while poorly informative about it, if it is also sensitive to other dimensions. One implication is that in a multi-dimensional world, stimulus properties such as ITD are optimally coded with heterogeneous neural populations.},
 author = {Brette, Romain},
 doi = {10.1121/1.3488311},
 issn = {1520-8524},
 journal = {The Journal of the Acoustical Society of America},
 keyword = {Sensory Receptor Cells},
 language = {eng},
 month = {Nov},
 number = {5},
 pages = {2965--2972},
 pmid = {21110592},
 title = {On the interpretation of sensitivity analyses of neural responses.},
 volume = {128},
 year = {2010}
}