Fig. 1.

An illustration of MVPA classification. (A) In this example, the analysis involves trying to classify two episodic-like memories from patterns of voxel activity in the hippocampus. Each memory is recalled five times, and the activity related to each recall trial is extracted. The string of letters represents activity from the hippocampus on each trial, labelled as either Memory A or B. The full dataset is split into a “training” set and a “testing” set, in this case assigning a single trial to the test dataset. Using the training set, an MVPA classifier is trained to differentiate memories A and B based on the patterns of activation in the hippocampus, and then tested on the test set. (B) In this example the test trial was classified as Memory B, which was a correct prediction. In a leave-one-out cross-validation, this process would be repeated ten times, each time leaving out a different trial as the test dataset. This cross-validation therefore yields a predicted label for every data trial in the analysis, which can then be compared against the real labels to produce an overall classification accuracy.