SCORE: Subject Coordinate Recovery for Label-Free Cross-Subject EEG-to-Image Retrieval
SCORE aligns EEG signals from new users without labeled calibration by recovering subject-specific coordinate directions.
The paper argues that subjects preserve similar concept relationships in EEG features but express them along different axes. SCORE trains on source subjects, then at deployment freezes the encoders, selects EEG-image landmarks, and estimates an orthogonal transformation for the target subject. In 200-way retrieval, it beats the unadapted baseline for every target subject on two public benchmarks. Reported Top-1/Top-5 scores are 53.23%/83.55% on THINGS-EEG2 and 12.01%/32.16% on Alljoined-1.6M. ArXiv · AI/CL/LG's note
The paper argues that subjects preserve similar concept relationships in EEG features but express them along different axes. SCORE trains on source subjects, then at deployment freezes the encoders, selects EEG-image landmarks, and estimates an orthogonal transformation for the target subject. In 200-way retrieval, it beats the unadapted baseline for every target subject on two public benchmarks. Reported Top-1/Top-5 scores are 53.23%/83.55% on THINGS-EEG2 and 12.01%/32.16% on Alljoined-1.6M. ArXiv · AI/CL/LG's note
score 4