Universal BCI Personalization: One API for Frozen EEG Trunks and Foundation Models
The paper argues the main value is a trunk-agnostic personalization interface for EEG models, not a new classifier trick.
Nimbus Personalizer puts a Bayesian head, with an optional affine mid-tier, on top of frozen EEG encoders so integrators can swap trunks without rebuilding personalization. The same surface is tested across five classical trunks, four motor-imagery datasets, and the REVE foundation encoder. The author says the head can recover much of fine-tuning’s accuracy gain where embeddings have enough capacity, with far lower adaptation time. Results are exploratory, with subject-level confidence intervals and no confirmatory tests. ArXiv · AI/CL/LG's note
Nimbus Personalizer puts a Bayesian head, with an optional affine mid-tier, on top of frozen EEG encoders so integrators can swap trunks without rebuilding personalization. The same surface is tested across five classical trunks, four motor-imagery datasets, and the REVE foundation encoder. The author says the head can recover much of fine-tuning’s accuracy gain where embeddings have enough capacity, with far lower adaptation time. Results are exploratory, with subject-level confidence intervals and no confirmatory tests. ArXiv · AI/CL/LG's note
score 4