Figure 4. Multi-channel recordings allow decoding of conscious percept on single trials.
(a) Left: Average responses (baseline-subtracted and normalized) of cells (rows) to 96 stimuli (columns) from six categories, including faces and other objects. Right: Waveforms of cells corresponding to rows on the left. Gray vertical bar on left indicates cells that significantly preferred face over object in the physical condition (p<0.05). (b) Top: Example eye trace across 24 trials as in Figure 1b during binocular rivalry (i.e., only perceptual, no physical switches). The inferred percept across trials according to eye trace is indicated by shading (red = face, blue = non face object). Small black dots on top of eye traces indicate time points where our method detected saccades (see Materials and methods), which are used in Figure 5. Bottom: Response time course snippet of a population of 81 neurons recorded with a Neuropixels-like probe in ML simultaneously to the eye trace at top. Each row represents one cell; ordering same as in (a). Face-selective cells indicated by gray vertical bar on left. (c) Normalized average population response across all significantly face-selective ML cells recorded from one Neuropixels session (same session as in a and b) to perceptual switch from object to face (red) and face to object (blue). Shaded areas indicate standard error mean across cells. (d) Cross-validated decoding accuracy of a linear classifier trained to discriminate trials of inferred percept face vs. inferred percept object for the physical switch condition (x-axis) and perceptual switch condition (y-axis). Each plus symbol represents a session of neurons recorded simultaneously with multi-channel electrodes.