Brain Encoding Models for Multimodal Movie fMRI Responses
A brain encoding model predicts recorded neural activity, here whole-brain fMRI responses parcellated into functional regions, from the stimulus that evoked it. This Idea concerns encoding models for multimodal naturalistic movies, whose inputs combine visual frames, audio, and language transcripts, and whose central quality criterion is generalization to out-of-distribution stimuli rather than fit within the training distribution, as formalized by the Algonauts Project 2025 Challenge on the CNeuroMod dataset.
2501.00504
The Algonauts Project 2025 Challenge is a community benchmark for building brain encoding models that predict whole-brain human fMRI responses to multimodal naturalistic movie stimuli. An encoding mo…