Back

Fail closed trust gated synthetic augmentation governs tail risk under subject shift in EEG

Choi, D.; Yip, C.; Choi, A.; Park, J.

2026-01-28 bioinformatics
10.64898/2026.01.26.701638 bioRxiv
Show abstract

Synthetic augmentation can silently harm subject-disjoint EEG generalization. We propose trustgated augmentation (TGA), a control layer that scores synthetic windows with a teacher trained on real data for label consistency and confidence; only samples above a confidence quantile q are eligible. A fail-closed selector injects synthetic data only if validation AUROC exceeds real-only by a margin, otherwise reverting to real-only. In PainMunich chronic-pain EEG (n = 189) at 5% subject scarcity, ungated augmentation harmed 56% of paired runs ({Delta}AUROC< -0.01), whereas TGA at q = 0.99 reduced harm to 24% with comparable mean AUROC. In BCI IV-2a motor imagery (n = 9) at 25% scarcity, strict gating improved AUROC (0.679 vs 0.627) and reduced harm (0.16 vs 0.44). A covariance-manifold audit showed synthetic windows were strongly off-manifold (mean distance ratio 2.39 x 104), motivating explicit governance.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

1
Nature Communications
4913 papers in training set
Top 2%
22.8%
2
Nature Machine Intelligence
61 papers in training set
Top 0.1%
14.5%
3
Proceedings of the National Academy of Sciences
2130 papers in training set
Top 13%
4.9%
4
Nature Methods
336 papers in training set
Top 2%
4.0%
5
Advanced Science
249 papers in training set
Top 5%
3.6%
6
Nature Medicine
117 papers in training set
Top 0.8%
3.6%
50% of probability mass above
7
PLOS ONE
4510 papers in training set
Top 38%
3.6%
8
Neuron
282 papers in training set
Top 4%
3.3%
9
PLOS Computational Biology
1633 papers in training set
Top 11%
3.1%
10
Nature
575 papers in training set
Top 8%
2.8%
11
Science Advances
1098 papers in training set
Top 12%
2.1%
12
Communications Biology
886 papers in training set
Top 6%
1.9%
13
Cell Systems
167 papers in training set
Top 6%
1.8%
14
Scientific Reports
3102 papers in training set
Top 57%
1.7%
15
Bioinformatics
1061 papers in training set
Top 7%
1.7%
16
eLife
5422 papers in training set
Top 45%
1.5%
17
Nature Computational Science
50 papers in training set
Top 0.7%
1.5%
18
npj Digital Medicine
97 papers in training set
Top 3%
1.2%
19
Nature Biomedical Engineering
42 papers in training set
Top 2%
0.9%
20
Patterns
70 papers in training set
Top 2%
0.9%
21
Nature Biotechnology
147 papers in training set
Top 7%
0.8%
22
Science
429 papers in training set
Top 19%
0.8%
23
Nature Neuroscience
216 papers in training set
Top 6%
0.8%
24
Nature Genetics
240 papers in training set
Top 8%
0.7%
25
NeuroImage
813 papers in training set
Top 6%
0.7%
26
Physical Review Research
46 papers in training set
Top 1.0%
0.7%
27
Genome Research
409 papers in training set
Top 5%
0.7%
28
Cancer Research
116 papers in training set
Top 4%
0.7%
29
The American Journal of Human Genetics
206 papers in training set
Top 5%
0.5%
30
npj Systems Biology and Applications
99 papers in training set
Top 3%
0.5%