(a) Decoding accuracy for an individual (participant 1) against number of encoding channels progressively incorporated into the decoding calculation, starting at the beginning of a list of channels sorted by the general linear model (GLM) weight value. The blue line shows the decoding accuracy validated in the scene choice (SC) sessions (sessions 1, 2, 3, 5, for training, and session 4 for test), and the red line shows the cross-task decoding accuracy for the motion detection (MD) task (the SC sessions for training, and the MD session for test). Dashed lines and dot-and-dash lines indicate decoding accuracies with the whole-scene model and the view-part model, respectively. (b) Different encoding schemes lead to different decoding performances in the SC task (left), where each decoding accuracy was examined in the validation session (session 4). Similar tendencies were also observed in the MD task (right). Black asterisks indicate significance (Wilcoxon signed rank test, p < 0.05). (a,b) Horizontal, black dotted lines indicate chance level (eight classes, 12.5%). (c) When the Hamming decoder was successful in decoding the scene prediction, the participants showed shorter RTs (blue histogram) than those in all decoding trials in MD (grey, skewness = 0.90, mode = −0.63). This RT reduction was most prominent in the optimal encoding model. The RTs were normalised to have mean zero and variance one within each participant.