Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster F in Poster Session F: Thursday, August 6, 1:45 – 3:30 pm, Kimmel Center, Shorin & Rosenthal Rooms

Neural Prediction Decorrelation in Human Auditory Cortex

David Skrill¹, Jenelle Feather², Samuel Victor Norman-Haignere¹; ¹University of Rochester, ²Carnegie Mellon University

Presenter: David Skrill

Model comparison is central to scientific progress in sensory neuroscience, yet distinct models often make similar neural predictions when evaluated on natural stimuli. We developed Neural Prediction Decorrelation (NPD), a method that synthesizes stimuli for which competing encoding models produce decorrelated predictions across an entire cortical population. We applied NPD to compare standard and adversarially robust deep neural network (DNN) models of fMRI responses in human auditory cortex, revealing a dramatic difference in prediction accuracy completely obscured when using natural sounds alone. We find that fMRI responses to NPD sounds occupy the same low-dimensional space as natural sounds, within which the models make highly divergent predictions. These results demonstrate that targeted stimulus synthesis can reveal model differences otherwise masked by traditional evaluation stimuli.

Topic Area: Methods, Tools, Theory & Neural Coding