Tag
This paper improves the Huth encoding pipeline for fMRI decoding and introduces fMRIFlamingo, a direct fMRI-to-text framework using Llama 3.2, but finds that decoding success is driven by the language prior rather than neural input.