Fig. 4: Differential category coding by individual modules.

a Category recognition task. Participants are shown images on a laptop screen and asked to categorize them as people, places, animals, or objects. During the 'PERSON' trials participants are shown pictures of famous people; due to licensing requirements we show a representative image that is not part of the stimulus set. b Response properties of individual modules. For an individual 'PERSON' trial the spike times are colored according to the module that the spike came from (left). The grayscale background represents the trial-averaged firing rate for each individual unit, and the colored lines show the average firing rate for each module. Neuronal spiking appears to cluster in time according to module membership. This was true for individual trials as well as on average across all 'PERSON' trials. c Coding of image category by individual modules. Each line represents the time-course of classification when using data from only that module. Some modules contain information about the category of the image (colored lines) whereas classification for others was below chance level (gray lines). d Schematic of within vs across module coding analysis. For the shuffle within, we shuffle unit identities using only units from that individual module before classifying a hold-out set, whereas for the shuffle across we shuffle the data only from units that belong to different modules. e Effects of shuffling on category classification. In this example, accuracy decreased when shuffling within, but remained well above chance level. In contrast, after shuffling across the classifier accuracy drops to near chance. f Scatter of classifier accuracy after shuffling within and after shuffling across for a single session. The shuffle within accuracy was significantly higher than the shuffle across. g Scatter of shuffle within and shuffle across accuracy for all sessions. Across all sessions, shuffle within accuracy is significantly higher than the shuffle across accuracy. Different colored points denote the individual arrays.