Wet-trained EEGNet
75.0% → 45.6%
29.4 percentage-point lossMoving the same fitted model from wet to dry recordings reduced balanced accuracy by 29.4 percentage points.
Sensor transfer · 102 participants
Dry vs. wet electrodes
Wet and dry recordings are not interchangeable inputs. The clearest mismatch appears when one EEGNet is trained on wet recordings and evaluated on dry recordings from held-out people—but sensor type is entangled with session order and time on task.
−29.4 ppWet-trained EEGNet: dry minus wet balanced accuracy
The largest observed cross-sensor loss
Across the same 102 people, the paired participant bootstrap interval is −34.5 to −24.3 pp. This is a descriptive transfer result from one training seed, not a pure causal estimate of electrode hardware.
Source-trained models
Train on one sensor, test on both
Six folds hold out 17 people at a time. Both sensor recordings from each held-out person stay out of source training. Every cell is participant-mean balanced accuracy on 12-target, two-second SSVEP.
| Model and training source | Test on wet | Test on dry |
|---|---|---|
| Spectral ridgetrained on wet · participant-disjoint | 40.3%36.5%–44.3% | 35.2%31.8%–38.8% |
| Spectral ridgetrained on dry · participant-disjoint | 39.7%35.9%–43.6% | 37.2%33.5%–41.0% |
| EEGNettrained on wet · participant-disjoint | 75.0%70.3%–79.5% | 45.6%40.5%–50.8% |
| EEGNettrained on dry · participant-disjoint | 64.0%59.1%–68.8% | 56.8%51.9%–61.6% |
Dry-trained EEGNet
56.8% → 64.0%
7.3 percentage-point gainTesting the dry-trained model on wet recordings increased balanced accuracy by 7.3 points; 95% interval +4.2 to +10.4 pp.
A different kind of reference
82.3% / 64.6%
64.6% dry FBCCA balanced accuracyFBCCA reads each target recording analytically. It does not learn from source participants, so it is not a transfer-trained model.
Analytic references
No source-training arrow belongs here
CCA, FBCCA and the power reference operate on the target recording. Listing them once per source sensor would duplicate identical predictions and imply training that never happened.
| Reference | Wet recording | Dry recording |
|---|---|---|
| Same-frequency poweranalytic or fixed reference | 39.8% · 95% interval 35.4%–44.3% | 36.0% · 95% interval 31.8%–40.2% |
| CCAanalytic or fixed reference | 81.6% · 95% interval 77.7%–85.1% | 65.5% · 95% interval 60.6%–70.2% |
| FBCCAanalytic or fixed reference | 82.3% · 95% interval 78.7%–85.7% | 64.6% · 95% interval 59.5%–69.6% |
Methods & limits
What this comparison can support
These are descriptive intervals over people under a fixed protocol. They are not confidence bounds for all headsets, sessions or model retraining choices.
Native wearable evidence
This dataset records both sensor types with a native eight-channel headband. It is separate from mobile results that select eight posterior scalp channels from a larger recording.
Order remains a confound
Wearing order and time on task are entangled with sensor type. Both order strata show the wet-trained drop, but the design cannot isolate hardware as the sole cause.
Amplitude units are unresolved
The release does not document a physical amplitude unit. The scale-dependent foundation-model runs remain held; no unit is guessed here.
One EEGNet seed
The source-trained EEGNet cells use one training seed. The paired intervals describe participant variation conditional on that fitted protocol.
An Open Dataset for Wearable SSVEP-Based Brain-Computer Interfaces
Zhu, F., Jiang, L., Dong, G., Gao, X., & Wang, Y. (2021). An Open Dataset for Wearable SSVEP-Based Brain-Computer Interfaces (Version 4) [Data set]. Figshare. https://doi.org/10.6084/m9.figshare.13560281.v4
Zhu, F., Jiang, L., Dong, G., Gao, X., & Wang, Y. (2021). An Open Dataset for Wearable SSVEP-Based Brain-Computer Interfaces. Sensors, 21(4), 1256. https://doi.org/10.3390/s21041256
Aggregate results computed from the official Tsinghua BCI Lab author mirror snapshot acquired 2026-09-20. Local checks do not establish byte identity with Figshare v4.
Dataset record ↗ · Sensors paper ↗ · CC BY 4.0
Data source: reviewed aggregate JSON · schema bci-report-public-deployment-topics-v1 · generated 2026-09-20.