Reproduce “LuMamba: Latent Unified Mamba for Electrode Topology-Invariant and Efficient EEG Modeling” 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
Danaé Broustail, Anna Tegon, Thorir Mar Ingolfsson, Yawei Li, Luca Benini
LuMamba handles electrode layouts and long EEG recordings by mapping channel information into a fixed representation and processing it with bidirectional Mamba blocks. The authors compare masked reconstruction, LeJEPA, and combined pre-training. On APAVA, mixed pre-training reaches mean AUPR 0.970 ± 0.012, above either objective alone, but reconstruction-only performs better on TUAR and TUSL; APAVA results are fine-tuned and include four test subjects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.