The links below show results from analyses in which we try to separate information about the first stimulus from information that is due to the second stimulus (as well as trying to separate information about the second stimulus from information due to the first stimulus). On nonmatch trials, each first stimulus was consistently paired with the same nonmatching second stimulus throughout the course of a recording session. This gives rise to a situation where if one knows what the second stimulus is on nonmatch trials with perfect accuracy, one would also know what the first stimulus is, and thus after the second stimulus is shown, above chance decoding of information about the first stimulus could be due to the classifier using information about the second stimuli (e.g., above chance decoding after 3000 ms into the trial in Figs. 1a, S3a, etc. could be due to either the first stimulus or the second stimulus). Below we describe an analysis where we try to separate information is coming from the first stimulus from information coming from the second stimulus.
In the spatial experiment, the pairing between first and second stimuli on trials was the same for all neurons (this pairing consisted of stimuli that were shown on opposite sides of the screen). Because match and nonmatch trials each occurred about equally, if we do a binary decoding analyses where we try to tell the first stimulus apart from its nonmatching pair, we should still be at chance (~50% correct) at the time when the second stimulus is shown if there is no information about the first stimulus at this time (this same method can also be used to separate information about the second stimulus from information about the first stimulus). The below links show the results from this binary decoding analysis between match/nonmatch pairs, averaged over all 4 pairs. Overall the results look qualitatively similar to the results in the paper that do not separate out the influence of the first and second stimuli.
A similar analysis was done in for the feature task, however things are slightly more complicated because the first-second stimulus pairing on nonmatch trials was different for different neurons. To get around this we did an analysis where we created regular pseudo-population vectors for the first stimulus, and we tried to decode these from pseudo-population responses to the stimuli from nonmatching paired trials (i.e., different stimuli gave rise to the responses to different neurons in the pseudo-population for the second stimulus, but these nonmatching stimuli were always consistently paired with the same first stimulus for any given neuron, again making chance 50%). The results in the links below are averaged over all 8 first-stimulus second-stimulus pairs (and an analogous analysis was done for the second stimulus). Overall the results look qualitatively similar to the results in the paper that do not separate out the influence of the first and second stimuli.