Neural integral operators improve fMRI encoding and decoding
A new study from arXiv (2605.20389) explores neural integral-operator models for fMRI encoding and decoding. The framework uses latent fixed point iterations in an auxiliary space to classify stimuli and predict brain dynamics. Tested on two open-source datasets, the research systematically compares short vs long temporal windows and visual cortex vs whole brain recordings, analyzing their impact on performance and latent-space geometry.
Key facts
- arXiv paper 2605.20389 investigates neural integral operators for fMRI
- Model performs fixed point iterations in an auxiliary space
- Evaluated on two open-source fMRI datasets
- Compares short and long temporal windows
- Compares visual cortex vs whole brain recordings
- Focuses on nonlocal spatiotemporal context
- Tasks include decoding stimuli and encoding fMRI dynamics
Entities
Institutions
- arXiv