<ul><li>Researchers have developed BraInCoRL, an in-context learning model using a transformer architecture to predict neural responses in the human higher visual cortex.</li><li>BraInCoRL aims to learn image-computable models without the need for individual-level, large-scale fMRI datasets, thus reducing the time and cost constraints associated with data acquisition.</li><li>The model outperforms existing voxelwise encoder designs in a low-data regime and exhibits strong test-time scaling behavior, showcasing its generalizability to new subjects and stimuli.</li><li>Additionally, BraInCoRL enhances the interpretability of neural signals in the higher visual cortex by focusing on semantically relevant stimuli and enables mappings from natural language queries to voxel selectivity.</li></ul>

Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex

Discover more