Abstract
Using Representational Similarity Analysis (RSA) and fMRI, we recently demonstrated evidence for individually unique visual object representations that predict perceived similarity (Charest et al. 2014). To date, little is known about how such idiosyncratic representations are established in the brain when perceiving an object. RSA analyses of MEG/EEG data provide novel opportunities to track brain representations with high temporal precision. To this end, we recorded 128-channel high-density EEG while participants (n=20) were presented with personally-meaningful and unfamiliar object images. These included pictures of bodies, faces, places, and man-made objects. We trained a Fisher linear discriminant on spatial features in order to classify each image pair. These pairwise decoding accuracies form entries in "representational dissimilarity matrices" (RDMs), which characterise the representational geometry at each time point. The average decoding performance that we observed using EEG is comparable to previous MEG results (Cichy et al. 2014), with significant image decoding from ~70ms post-stimulus onset and peak decoding at ~150ms. To measure the reliability of representational geometries, we computed the RDM replicabilities (correlation between lower-triangular entries) across data-splits at each time point. We observed replicable RDMs from ~80ms post-stimulus onset which carried over a sustained time window. Comparing dissimilarity matrices at each time-point within and between individuals revealed the neural dynamics of individually unique object representations. Importantly, we found earlier individuation effects for unfamiliar compared to personally-meaningful images. This is consistent with our previous fMRI results where an individuation effect was observed in early visual cortex only for unfamiliar images. In summary, successful image decoding suggests rich spatial information contained in EEG topographies characterising stable object representations. Furthermore, harnessing the temporal precision offered by EEG, we reveal an interaction between early visual processes and personally meaningful objects. Using MVPA applied to EEG enables to characterise the build-up of idiosyncratic object representations.
Meeting abstract presented at VSS 2017