3/7 This transformation had direct functional consequences: linear classifiers decoded odour identity more accurately from output than input activity.
By computationally transforming the representational geometry, we found that improved decoding accuracy was directly linked to decorrelation itself.
