Decoding including and excluding neurons that have temporal
trends over trials
One
of the preprocessing stages in the readout analysis was to remove all
neurons that showed temporal trends over trials. Neurons with
nonstationary trends were defined as those neurons that had an average
firing rate variance in 20 consecutive trials that was less than half
the trial firing rate variance over the whole session. Because
the
stimuli were presented in random order, the average variance in 20
tirals should be roughtly equal to the variance over the whole session
(and deviations from this suggest that there could be artificats in the
firing rate due such factors are movement in the recoding electrode).
42 ITC and 34 PFC neurons met this trend criterion, and were
excluded from the decoding analyses in the paper. Below shows
the
results for decoding identity and category informatoin from ITC and PFC
when these trends neurons were both included and excluded. As
can
be seen, the decoding accuracies are usually slighly lower when the
neurons with trends are included, however overall the pattern of
results is very similar. Also shown below are 4 random ITC
and 4
random PFC neurons that were considered 'trend neurons' as defined by
the criterion listed above (in order for the reader to get a sense of
what the trends in the firing rates looked like).
Home