Figure 1.

Decoding stimulus perception, stimulus retention, motor selection, and stimulus comparison from hippocampal ensemble activity. A: The decoder classifies binary labels--low and high tones during sample and delay periods, no-go and go responses, and match and non-match tones during test periods--based on neuronal ensemble activity patterns. The dots show the raw spiking activity pattern of the hippocampal ensemble obtained while the rat performs the DNMS task. The circles are linear classifiers, each of which linearly combines the spiking activities as inputs (weighted sum plus bias; bias component not shown). The weights are determined using a statistical learning algorithm (linear support vector machine (SVM)) applied to each training dataset such as those shown in B-G. Finally, each SVM specialized for stimulus perception (SVM H/L P), stimulus retention (SVM H/L R), motor selection (SVM G/NG), and stimulus comparison (SVM M/NM), classifies the behavioral events. B-G: Typical raster plots of raw spiking activity of hippocampal ensemble #1 used for SVM classification in low tone, high tone, no-go response, go response, non-match, and match trials, respectively. The middle traces show sequences of events within a trial. The lower lines show intervals in which SVM used ensemble activity to classify low and high tones during the sample (solid lines in B and C) and delay (dotted lines in B and C) periods, or no-go and go responses or non-match and match tones during test (solid lines in D-G) periods.