Color Space Geometry Uncovered with Magnetoencephalography

Abstract

Highlights d Stimulus color can be decoded from surface magnetoencephalography (MEG) recordings d Perceptual representations give rise to semantic representations, but not the reverse d The results reveal a neural geometry of color space that is dynamic d The geometry explains universal color-naming patterns and generates new hypotheses

Temporal dynamics of the neural representation of hue and luminance polarity

“…While the results show that representations of hue and luminance contrast are somewhat decoupled, they also confirm prior work showing the existence of representations that combine hue and luminance contrast (Rosenthal et al, 2020b): decoding performance for the identity problems, in which classifiers were trained using responses to stimuli that were distinguished by a combination of luminance contrast and hue, was always better than decoding performance for the generalization problems, in which classifiers were trained using responses to stimuli that were only distinguishable by one dimension, invariant to the other ( Figure 2 ). These results are therefore consistent with the idea that in addition to separate representations of hue and luminance contrast, the visual system has representations that combine hue and luminance contrast, perhaps reflecting the joint selectivity for hue and luminance contrast evident in some neural populations in the LGN(Wiesel and Hubel, 1966) and V1(Gegenfurtner, 2003; Johnson et al, 2008; Horwitz and Hass, 2012).…”

“…Note that decoding using pupil diameter and eye position cannot account for decoding of color (Rosenthal et al, 2020b).…”

“…Color can be decoded from MEG activity(Hermann et al, 2015; Rosenthal et al, 2017; Sandhaeger et al, 2019; Teichmann et al, 2019; Hajonides et al, 2020; Hermann et al, 2020; Rosenthal et al, 2020a). We have previously shown that MEG data elicited by colors provide evidence of joint coding of hue and luminance contrast that reflect color-naming patterns and can be used to construct a geometry of the neural representation of color (Rosenthal et al, 2020b). In the present work, we use multivariate analyses of the same large dataset to take up two unresolved issues: first, we test the extent to which hue and luminance contrast are encoded independently and whether these representations generalize across time; and second, we exploit the exquisite temporal resolution of MEG to tease apart the time course of the representations for hue and luminance contrast (the dataset is available here: https://doi.org/10.18112/openneuro.ds003352.v1.0.0; and here: https://neicommons.nei.nih.gov/#/).…”

“…Here we explore a new approach to address the neural mechanisms for encoding hue and luminance contrast, using magnetoencephalography (MEG) in humans, coupled with multivariate analysis 44–48 ( Figure 1 ). Color can be decoded from MEG activity 49–56 . In the present work, we exploit the exquisite temporal resolution of MEG to tease apart the neural mechanisms for hue and luminance contrast.…”

Decoding of EEG signals reveals non-uniformities in the neural geometry of colour

“…To achieve this, we chose unique and non-unique hues that were maximally distant in a perceptual space -red and green, orange and turquoise (see details of the stimulus set in Methods). As already reported by Rosenthal et al (2021) and Hermann et al (2021), inter-hue differences in decoding efficiency manifest even between such evenly spaced colours. To better understand the non-uniformity of this neurometric colour space, in our next experiment we aimed to investigate the structure of the decoding manifold by introducing proximal neighbours, clockwise and counterclockwise to each hue.…”