Open-source PyTorch model GRFNet targets multichannel time-series classification
Drawing on MSc EEG decoding work, a developer open-sourced GRFNet, a lightweight PyTorch model combining random convolutional kernels with learned spatial filters. The default 22-channel, 4-class configuration has 3,847 parameters and accepts any input length while adapting to sampling rate. Subject-independent evaluation on EEGMMIDB is underway; no benchmark numbers yet.