SpecBraM: What Should an EEG Foundation Model Predict? Masked Band-Power Prediction versus Waveform Reconstruction

Peng Xie, Yequan Bie, Jianda Mao, Kani Chen (Hong Kong University of Science and Technology)

Summary

A controlled study of the masked-modeling target. Same 12-layer criss-cross transformer, data and step budget; the only change is what masked patches predict. Instead of reconstructing the raw waveform (CBraMod, REVE) or VQ codes (LaBraM), SpecBraM predicts log energy in five fixed frequency bands per masked channel-time patch, dropping phase that can't be inferred from neighbours.

Reported benchmarks