Paper notes / a reading log, in progress.
BCI papers I have been working through while studying the field. Click a paper to read the note. Longer write-ups on eeg-mi-benchmark phases coming to Substack and Bluesky post-CNEW.
FIG · 02 — Motor imagery trial · C3 ERD during right-hand imagery · 5ch · simulated
- 2002
Brain-computer interfaces for communication and control
Wolpaw et al.Clinical NeurophysiologyFoundationsReview“The field's origin paper. Every BCI paper traces back here. Defines the signal acquisition to application pipeline that everything else builds on.”
- 1973
Toward direct brain-computer communication
VidalAnnual Review of Biophysics and BioengineeringFoundationsHistory“The paper that coined "BCI." Short and readable. The core challenges Vidal named in 1973: SNR, real-time processing, user training. Still the core challenges now.”
- 1999
Event-related EEG/MEG synchronization and desynchronization: basic principles
Pfurtscheller & Lopes da SilvaClinical NeurophysiologyERD/ERSNeurophysiology“Explains why the 8–30 Hz bandpass works and what mu/beta suppression over C3/C4 actually means physiologically. The theory behind every motor imagery plot I've made.”
- 2008
Optimizing spatial filters for robust EEG single-trial analysis
Blankertz et al.IEEE Signal Processing MagazineCSPFeature Extraction“CSP: simultaneous diagonalization of two covariance matrices. Implemented from scratch to make sure I understood it before using MNE's version. Phase 1 classifier in eeg-mi-benchmark.”
- 2008
Filter bank common spatial pattern (FBCSP) in brain-computer interface
Ang et al.IEEE IJCNNCSPFeature Selection“FBCSP extends CSP across frequency bands and selects features via mutual information. Won BCI Competition IV 2a. The standard baseline everything new gets compared against.”
- 2012
Review of the BCI Competition IV
Tangermann et al.Frontiers in NeuroscienceBenchmarksMotor Imagery“Covers BNCI2014001, the dataset in eeg-mi-benchmark. Understanding the competition protocol is what makes a benchmark number mean something.”
- 2018
A review of classification algorithms for EEG-based BCIs: a 10-year update
Lotte et al.Journal of Neural EngineeringRiemannianReview“The Riemannian geometry section is the key part. Covariance matrices on a manifold, not Euclidean space. Consistent cross-subject generalization with less tuning than deep learning.”