Minimal criteria for a sensory encoding model
We identify three criteria that any encoding model of a sensory cortical system should meet:
Stimulus-computability: The model should accept arbitrary stimuli within the general stimulus domain of interest;
Mappability: The components of the model should correspond to experimentally definable components of the neural system; and
Predictivity: The units of the model should provide detailed predictions of stimulus-by-stimulus responses, for arbitrarily chosen neurons in each mapped area. These criteria may sometimes be in tension—insisting on mappability at the finest grain might hinder identifying models that actually work for complex real-world stimuli, since low-level circuit tools may operate best in reduced stimulus regimes. While seeking detailed models of neural circuit connectivity in simplified contexts is important, if such models do not add up in the aggregate to accurate predictors of neural responses to real-world stimuli, the utility of their lower-level verisimilitude is limited.
see also
Type:
Tags:
Status:
Location:
Created: 21-11-24 15:35
Source
Yamins, D. L. K., & DiCarlo, J. J. (2016). Using goal-driven deep learning models to understand sensory cortex. Nature Neuroscience, 19(3), 356–365. https://doi.org/10.1038/nn.4244