Fig. 7

Comparing fixed- and flexible-readout linking models. a–c Same conventions as Fig. 2a, b, except curves plot: a the fit of the passive cortical response with a fixed readout, b the cortical response during task performance with a fixed readout, and c the cortical response during task performance with a flexible readout. The fixed-readout model forces any change in perceptual sensitivity to be the result of differences in sensory response between tasks by using only a single set of cortical readout weights (eight weights and one bias term). The flexible-readout model allows a different set of weights for each task condition (16 weights and 1 bias term). d Tjur’s CD for the models in a–c. Averages are shown as a bar on the axis. e Conventions as in d for the two-area models with only V1 and MT. f Model comparison of the cross-validated likelihood ratio (difference in log-likelihood) between the task-response fixed-readout and task-response flexible-readout models. Evidence for the fixed-readout model is plotted to the left and flexible-readout to the right (none of the fits are in favor of the fixed-readout model). g As in f for Tjur’s CD. Markers in a–c indicate the average across observers. Markers in d and e indicate individual observers. Error bars are the 95% confidence intervals. Some error bars are hidden by the markers