A new study has audited the topographic organization of the TRIBE v2 multimodal brain-response model, finding that it accurately predicts motion selectivity in area MT+ but fails to replicate retinotopy and tonotopy. The model's video features are averaged spatially, which discards visual-field position information necessary for retinotopy. Similarly, its predictions for auditory cortex show a very narrow range of best frequencies, indicating a lack of tonotopic gradient. While the model shows promise for experiments relying on feature identity, it is not supported for those dependent on visual-field or frequency position. AI
IMPACT This research highlights limitations in current multimodal models for representing spatial and frequency-based information, suggesting areas for future development.
RANK_REASON Research paper published on arXiv detailing findings about a multimodal brain-response model. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →