An In-Silico Audit of Topographic Organization in a Multimodal Brain-Response Model
Read the original on arXiv AI →The study evaluates TRIBE v2, a multimodal brain‑response model that predicts fMRI activity from video, audio, and text. Using mapping stimuli and a fitting pipeline, the authors find that TRIBE v2 fails to reproduce classic topographic maps—no contralateral retinotopy, no polar‑angle map, and no tonotopic gradient—yet it accurately captures motion selectivity in area MT+. These results highlight limitations in the model’s spatial and frequency encoding while confirming its utility for feature‑identity‑based analyses.
Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.