Related Experiment Videos
Narrow versus wide tuning curves: What's best for a population code?
A Pouget1, S Deneve, J C Ducom
1Georgetown Institute for Cognitive and Computational Sciences, New Research Building, Room EP04, 3970 Reservoir Road NW, Washington DC 20007, USA. alex@giccs.georgetown.edu
Neural Computation
|February 9, 1999
Summary
Sharpening neural tuning curves may not always improve coding quality. The optimal curve width depends critically on noise characteristics, challenging common assumptions in neurophysiology.
Area of Science:
- Neuroscience
- Computational Neuroscience
- Systems Neuroscience
Background:
- Neurophysiologists evaluate neural code quality for sensory/motor variables.
- A common belief is that sharpening tuning curves improves code quality, but only to a certain extent.
Purpose of the Study:
- To re-evaluate the common belief regarding the impact of tuning curve sharpening on neural code quality.
- To investigate the role of noise covariance in determining optimal tuning curve properties.
Main Methods:
- Analysis of neural coding models.
- Theoretical investigation of tuning curve properties under different noise assumptions.
- Examination of the impact of tuning curve manipulations (e.g., sharpening, gain increase) on information transmission.
Main Results:
- The belief that sharpening tuning curves universally improves code quality is challenged.
- The optimal tuning curve width is critically dependent on the covariance of neural noise.
- Improper assumptions about noise can lead to flawed conclusions about code quality.
Conclusions:
- In the general case, it cannot be determined whether narrow or wide tuning curves are superior.
- The optimal tuning curve profile is contingent upon the specific noise characteristics, particularly its covariance.
- Conclusions drawn from tuning curve manipulations must account for underlying noise properties.