Decoding subjective simultaneity from neuromagnetic signals

The present study examined neural correlates of subjective simultaneity by using magnetoencephalography. Observers were asked to judge whether the visual and auditory stimuli occurred simultaneously. The subjective judgment for 90-ms-asynchronous stimuli showed trial-by-trial variation, and we successfully classified subjective simultaneity using neuromagnetic signals. We submitted raw MEG signals, a wavelet transform, and nonlinear dynamics to a naive Bayes classifier. In the case of raw signals and nonlinear dynamics, the classifier trained with the VA (where the visual stimulus was given first) or AV (where the visual stimulus was given second) data could predict the subjective simultaneity of the other VA (or AV) data at a rate better than chance. The classification rate using nonlinear dynamics was comparable to that using raw signals, despite the fact that the dimension was considerably low (101 vs. 88,000 dimensions). In the case of the wavelet transform, the classifier trained with the VA data was able to decode the AV data, and vice versa. These results suggest that (1) we can decode subjective simultaneity using MEG signals, (2) nonlinear dynamics may encode simultaneity specific to the order of the audiovisual inputs, (3) the time-frequency characteristics of neural activity may predict subjective simultaneity independently of the physical order of the audiovisual inputs, and (4) the neural activity (time-frequency characteristics) reflecting subjective simultaneity may share a common mechanism among different sensory modalities.