Fig. 3. Identifying sensory stimulus input units (vertices) of the ENN using an fMRI decoding analysis.

a We identified the sensory stimulus representations in empirical data using fMRI pattern decoding of stimulus activations. This corresponded to the sensory input component of the Guided Activation Theory. To decode visual features (i.e., color and orientation stimulus features) we decoded the vertices within each parcel in the visual network using a recent functional network atlas⁶⁵. To decode auditory features (i.e., pitch and continuity) we decoded the vertices within each parcel in the auditory network (see Methods). b The ENN sensory units, which were derived from a mask of the vertices that could successfully decode stimulus features (panels c–f). c Decoding of color features using task activation estimates (from a task GLM) during the stimulus presentation period of the C-PRO task. Chance was 25%; cortical maps were thresholded using an FDR-corrected threshold of p < 0.05. d Four-way decoding of orientation features. e Four-way decoding of auditory pitch features. f Four-way decoding of auditory continuity features.