Vin Bhaskara @vinbhaskara.github.io · Feb 28 @apurvaratan.bsky.social et al. (Nature 2021) collected neural responses from fMRI scans on a limited dataset of visual stimuli and subsequently trained a CNN model to replicate the activations. To their surprise, the CNN generalized to novel stimuli. (5/10)
0 likes 1 replies
?
Replies
The trained CNN model allowed for validating the specificity hypothesis at scale across a much larger dataset of images with a higher diversity of scenes. Interestingly, the group couldn't falsify their hypothesis that processing faces involved a specific region! (6/10)