Research questionHow can MEG foundation models become reusable across subjects, sites, and tasks despite limited training data?MEG foundation-model development is constrained by a small number of models, modest pretraining corpora, and limited benchmarks. These constraints make it difficult to determine which design choices support transfer across people, recording sites, tasks, and clinical settings. Latest papersRecent research connected to this question, newest first.A Roadmap for MEG Foundation ModelsThe source surveys MEG-specific and multimodal models, sensor- and source-space representations, sensor-geometry encoding, architectures, self-supervised objectives, pretraining data, transfer from EEG and generic time-series models, and integration with other modalities. It provides a field roadmap rather than evidence from a single controlled model comparison, while also discussing dataset reuse, evaluation, consent, privacy, access, and governance.research paper · Sep 3, 2026