Reproduce “Logit Averaging During Training Sets a New State of the Art for Non-Invasive Speech BCI” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Miran Özdogan, Gereon Elvers, Michael Bronstein, Oiwi Parker Jones
The authors introduce training-time logit averaging for classifying repeated MEG recordings: a shared model processes each trial, and averaged class scores receive one group-level loss. Across seven evaluated settings, it leads six, with imagined-speech MegNIST as the exception. On LibriBrain phoneme classification, it reaches balanced accuracy 0.641 versus 0.467 for the PNPL winner on the public test split, not hidden holdout labels.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.