brainnn

Neuromodulator-Conditioned BCI Decoding

Specs

Status: Research prototype (July 2026, active)
Role: Solo — architecture, implementation, evaluation, communication
Result: 40% zero-shot cross-subject accuracy on BNCI2014_001 (4-class motor imagery, chance = 25%)
Stack: Python · PyTorch · MNE · MOABB · Streamlit · Plotly
Novelty: Independent literature review confirmed no prior work conditions a BCI decoder on a 3-D neuromodulator estimate
Review: Architecture personally reviewed by Peter Dayan (MPI Tübingen)