arXiv:2606.06345v1 Announce Type: cross Abstract: Brain decoding is limited by the availability of labeled neural data, and remains challenging in low-data regimes. To address this issue, we investigate whether and when brain decoding can be boosted by augmenting small fMRI datasets with synthetic data generated by a pretrained model of fMRI responses to stimuli. We use TRIBE v2, a large encoding model pretrained on more than 1000 hours of fMRI responses to video, audio and language. For each dataset, we evaluate systematic grids that show how the performance of image decoders varies with the
Advances in large-scale AI models, as exemplified by TRIBE v2, are enabling new avenues for data augmentation in fields like neuroimaging, addressing longstanding data scarcity challenges.
This research shows a pathway to significantly improve brain decoding capabilities, which could accelerate the development of brain-computer interfaces, medical diagnostics, and potentially enhance AI's understanding of human cognition.
The ability to effectively augment limited fMRI data with synthetic data generated by sophisticated AI models changes the constraints on developing robust brain decoding systems.
- · Neuroscience researchers
- · Brain-computer interface developers
- · AI model developers
- · Medical diagnostics companies
- · Research relying solely on vast, difficult-to-acquire human fMRI datasets
- · Traditional fMRI data collection methodologies
This significantly improves the performance of brain-to-image decoding models, even with sparse real fMRI data.
Enhanced brain decoding could accelerate development of more intuitive and powerful human-computer interaction methods.
Advanced understanding of brain activity might lead to breakthroughs in treating neurological disorders or developing advanced cognitive AI systems.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG