Real-Time REM Sleep Stage Classification Using Deep Neural Networks and Multiple Task Learning
Bibliographic record
Abstract
Rapid Eye Movement (REM) sleep plays a critical role in memory consolidation, emotional regulation, and cognitive functioning. The accurate detection of REM sleep is essential to the diagnosis of sleep disorders and assessing the quality of sleep. We provide a comparative study of convolutional neural networks for classifying REM sleep stages in real-time using data from a commercially available smartwatch. Heart rate and acceleration measures are collected and consolidated into 2minute intervals for input to the model. Ground truth data has been gathered through Apple's HealthKit dataset. We statistically prove that Apple's HealthKit data is sufficiently close to the industry standard Cerebra Sleep Study System and is more affordable. We employ 5-fold cross-validation to evaluate neural network architectures tailored to individual users, aiming to enhance model performance. By integrating advanced features and incorporating Multi-Task Learning (MTL) techniques, we achieve significant improvements in real-time REM sleep detection, surpassing the benchmarks we established before this experiment. Our work focuses on creating personalized models for each user, which is why we initially work with a small dataset to test how the model performs and detect different sleep stages in real time. Our results indicate that the single-layer convolutional model with batch normalization and a kernel size of 3, as well as the single-layer convolutional model with max-pooling and a kernel size of 9, have superior performance. The best-performing model, developed for a single subject, achieved an Area Under the Curve (AUC) of 0.901, a correlation of 0.697, REM accuracy of 0.853, an F1 score of 0.708, a True Positive Rate (TPR) of 0.788, and True Negative Rate (TNR) of 0.881 on the test dataset.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".