There are a ton of seemingly different ways/people/conceptual-frameworks that all are influenced or related to :
- traditional computational neuroscience [trial-to-trial response variability (same stimulus) ⟶ this variability is NOT independent across neurons. it affects the coding capacity of a population of neurons]
- it is possible that there is some "undesired" response (cognitive/task/response/head-jerk/emotion) that is reliably linked to the stimulus, in which case it is technically a SIGNAL and not a noise correlation
- resting-state fMRI [the "assumption" is that there is no stimulus/cognitive/motor "events"; hence, the entire data is "noise"] [biomarkers, cortical parcellation, network neuroscience, Bayesian sampling]
- approaches for denoising [since correlation implies "structure" which means not random → this is an opportunity to analyze that structure and remove it]
- RSA [some multivariate (multivoxel) response analysis → your results are going to be influenced by noise correlations]
- functional "connectivity" [if you correlate two units from the same dataset, the results may partially or be dominated by noise correlations]
- traditional "voxel-wise" encoding analysis does not specifically try to model/characterize noise correlations
- pattern classification MVPA [the performance of your method is going to depend on the nature of the noise correlations]
- could be present in anatomical measurements!
There are many sources of noise, some neural (and interesting), some measurement related.
What about signal correlations?
- Working definition: is the noiseless signal measured from different brain units correlated?
- From a measurement point of view, often we try to work to get rid of the idea that it's because of sub-optimal measurement that the signals are correlated