Brain Computer Interfaces · Machine Learning
NeuroForge
Train a left-vs-right hand motor-imagery BCI on a virtual EEG subject. Tune the subject and the signal pipeline, watch the cursor respond trial by trial, and ask the AI copilot how to improve it.
Closed-loop session · trial 1 / 100
Fp1
Fz
C3
Cz
C4
Pz
O1
O2
0 s2 s
Cue: imagine left hand
Decoded: right — miss
Running accuracy 0% · effective ERD 0.41
Calibration (5-fold CV)
75%
Online accuracy
76%
ITR (Wolpaw)
4.1 bits/min
ERD start → end
0.41 → 0.32
Accuracy per block of 20 trials
chance 50%
60%
85%
75%
80%
80%
Block 1Block 2Block 3Block 4Block 5
Calibration features
● right-hand imagery · ● left-hand imagery · dashed: LDA boundary
BCI copilot
The scan tests every spatial filter and band on the calibration data. The AI copilot reads the results and suggests what to change.