Search Papers | Poster Sessions | All Posters
Poster B153 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Relational neural control: a method for investigating functional relationships across the brain
Alessandro T. Gifford1 (), Maya A. Jastrzębowska1, Johannes J. D. Singer1, Radoslaw M. Cichy1; 1Freie Universität Berlin
The functional relationships between early- and mid-level retinotopic regions of interest (ROIs) of the human visual cortex are not entirely understood. We address this gap by introducing Relational Neural Control (RNC), a neural-control-based method that jointly controls the activity in multiple ROIs by selecting images that align or disentangle their responses. We applied RNC on retinotopic visual cortex using the Neural Encoding Dataset (NED), a massive dataset of synthetic fMRI responses to naturalistic images. RNC found stimulus images that significantly aligned or disentangled both univariate and multivariate responses between retinotopic areas, and these controlling images contained interpretable visual patterns. Our contributions are threefold. First, we provide new quantitative and qualitative findings on functional similarities and differences across retinotopic areas. Second, we introduce RNC as a generalist method for controlling neural responses and uncovering functional relationships across the brain. Third, we release NED with tutorials, hoping it will boost research in cognitive computational visual neuroscience.
Keywords: method and data release neural control encoding models retinotopic visual cortex