Fig. 3: ROC curves and sample scaling results.
From: Weak supervision as an efficient approach for automated seizure detection in electroencephalography
We plot a median ROC curves when evaluating our model trained on 25,386 weak labels over 60-s and 12-s EEG clips for pediatric patients and b median ROC curves when evaluating our model trained on 32,596 weak labels over 60-s and 12-s EEG clips for adult patients. We also show model performance versus the number of weak labels used for training over pediatric patients (c) and adult patients (d) over the 12-s EEG clips. Error bars represent 95% confidence intervals from five training runs with different random seeds.
