DeepWaveBEHAVIOR EEG
VideoAI behavior labels (training sources)
Loading…
MULTI-SESSION MODEL BUILDER
Temporal Binary EEG Model Laboratory
Compare Gradient Boosting, supervised CNN 1D, logistic L2, and temporal waveform templates on independent binary detectors. No 15-class training.
02 / MODEL TRAINING
14 Binary Behavior Detectors — Select Algorithm
Six short-event and eight sustained-behavior detectors · independent nested cross-validation and optimized thresholds.
Select recordings in Old Sessions to start.
Filter continuous EEG before extracting windows. Family A: 0.1–6 Hz morphology, no EMG input; Family B: slow 0.1–6 Hz + 6–20 Hz, alpha 8–13 Hz, EMG 20–80 Hz and EMG envelope modulation. Offline zero-phase notch 50 Hz, 250 ms gap guard, 5 s filter-edge guard. Thresholds are tuned using only inner training sessions.
CNN 1D learns convolution kernels end-to-end from 64 time bins (A: slow 0.1–6 Hz; B: slow + EMG/alpha envelopes). Template matching uses the same filtered waves. CNN training may take significantly longer, especially with many sessions. All models use the same session-held-out validation and individual threshold tuning.
Ready
03 / INDEPENDENT EVALUATION
Binary Detector Comparison — Held-out Sessions
Each detector is trained independently. Thresholds are chosen on inner sessions, frozen and applied to the outer held-out session. Avoid duplicate raw EEG sessions.
EXPERIMENTAL 0.8.06
One algorithm at a time — 14 independent detectors
Choose an algorithm above and Build Model. Results use VideoAI-anchored candidate windows, not validated continuous event detection. Compare saved results across runs.
Research-only. No change to the active real-time classifier. Rare classes and sessions with inadequate examples are not scored.
01 / SELECTED SESSIONS
Dataset from Old Sessions 0
All session types are supported here. Only saved, accepted VideoAI labels count as teachers. Quality percentages refer to raw EEG, not display interpolation.
Reading selected sessions…
| Session | VideoAI | Behavior QC | Cognitive QC | Sync |
|---|
Dataset preview
COGNITIVE EEG
Cognitive labels (reference only — not trained)
Loading…
Teacher labels are estimates derived from VideoAI, not independently measured attention or behavior ground truth. EEG windows used for model training exclude visual-only interpolations.