Research map / Label-efficient and robust learning / OOD and anomaly detection
PSDNorm: Temporal Normalization for Deep Learning in Sleep Staging
Theo Gnassounou, Antoine Collas, Rémi Flamary, Alexandre Gramfort
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace BatchNorm with PSDNorm in first three conv layers | U-Sleep · LODO, all subjects, mean over subjects | balanced accuracy: 78.14 -> 79.51 | Improved |
| replace BatchNorm with PSDNorm | U-Sleep · LODO, balanced@400, mean over subjects | balanced accuracy: 77.22 -> 78.85 | Improved |
| replace InstanceNorm with PSDNorm | U-Sleep · LODO, all subjects, mean over datasets | balanced accuracy: 78.97 -> 79.15 | Improved |
| replace TMA preprocessing with in-network PSDNorm | U-Sleep · LODO, all subjects, mean over subjects | balanced accuracy: 78.77 -> 79.51 | Improved |
| replace BatchNorm with PSDNorm | U-Sleep · CHAT left out, balanced@400 | balanced accuracy: 66.55 -> 70.28 | Improved |
| use LayerNorm instead of BatchNorm | U-Sleep · LODO, balanced@400, mean over datasets | balanced accuracy: 77.55 -> 75.05 | Hurt |
| use barycenter target instead of whitening in PSDNorm (ablation) | PSDNorm · LODO, balanced@400, mean over datasets | balanced accuracy: 77.79 -> 78.34 | Improved |
| use PSDNorm instead of BatchNorm in CNNTransformer | CNNTransformer · LODO, balanced@400, mean over datasets | balanced accuracy: 76.38 -> 77.83 | Improved |
| use TMA preprocessing in CNNTransformer instead of BatchNorm | CNNTransformer · LODO, balanced@400, mean over datasets | balanced accuracy: 76.38 -> 76.30 | No clear effect |
| increase PSDNorm filter size f from 5 to very large (e.g., 21) (ablation) | PSDNorm · U-Sleep balanced@400 | balanced accuracy | Hurt |
| use PSDNorm filter size f in 5-11 range (ablation) | PSDNorm · U-Sleep balanced@400 | balanced accuracy | No clear effect |
| replace BatchNorm with PSDNorm (f=5) in training time | U-Sleep · training time per epoch | seconds per epoch: 161.94 -> 172.85 | Hurt |
Closest papers by meaning
- Generalizable Sleep Staging via Multi-Level Domain Alignment (AAAI 2024 · similarity 0.87)
- Personalized Sleep Staging Leveraging Source-free Unsupervised Domain Adaptation (AAAI 2025 · similarity 0.84)
- SleepFM: Multi-modal Representation Learning for Sleep Across Brain Activity, ECG and Respiratory Signals (ICML 2024 · similarity 0.84)
- Robust Sleep Staging over Incomplete Multimodal Physiological Signals via Contrastive Imagination (NeurIPS 2024 · similarity 0.82)
- SPDIM: Source-Free Unsupervised Conditional and Label Shift Adaptation in EEG (ICLR 2025 · similarity 0.82)
- Impact of Layer Norm on Memorization and Generalization in Transformers (NeurIPS 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
