Skip to main content
. 2020 Oct 27;11:5418. doi: 10.1038/s41467-020-18946-z

Fig. 3. Task paradigm and behavioral data.

Fig. 3

a Experiment web interface for our human categorization task. Participants categorized each image from an order-randomized circular array of the CIFAR-10 labels. b Examples of images and their human guess proportions. For many images (upper plane and cat), choices are unambiguous, matching the CIFAR-10 labels. For others (lower boat and bird), humans are far less certain. b The number of images by the entropy of their associated human guess distributions (less entropy means humans had greater consensus). d The confusion matrix for categories across all human judgments. Source data are provided as a Source Data file.