Figure 9.

Coding dynamics of pure and linearly selective cells (CS and LMS) versus NMS cells. Linear kernel SVM decoders were trained to perform cross-temporal decoding with different selectivity populations in the delay2 period, for both spatial and feature tasks, as indicated by the y axis. The decoder was then required to predict whether a match or nonmatch occurred at each time point based on a different test set of data, as indicated by the x axis. Normalized decoding accuracy is indicated in the color bar, demonstrating how spatial and feature WM representations can be decoded from specific patterns of neural activity. Coding of matching information for NMS cells is more stable across time for the spatial task.