AI Paper Atlas

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

ICLR 2026 · Label-efficient and robust learning › OOD and anomaly detection · approach: Uncertainty and misclassification detection · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
replace BatchNorm with PSDNorm in first three conv layersU-Sleep · LODO, all subjects, mean over subjectsbalanced accuracy: 78.14 -> 79.51Improved
replace BatchNorm with PSDNormU-Sleep · LODO, balanced@400, mean over subjectsbalanced accuracy: 77.22 -> 78.85Improved
replace InstanceNorm with PSDNormU-Sleep · LODO, all subjects, mean over datasetsbalanced accuracy: 78.97 -> 79.15Improved
replace TMA preprocessing with in-network PSDNormU-Sleep · LODO, all subjects, mean over subjectsbalanced accuracy: 78.77 -> 79.51Improved
replace BatchNorm with PSDNormU-Sleep · CHAT left out, balanced@400balanced accuracy: 66.55 -> 70.28Improved
use LayerNorm instead of BatchNormU-Sleep · LODO, balanced@400, mean over datasetsbalanced accuracy: 77.55 -> 75.05Hurt
use barycenter target instead of whitening in PSDNorm (ablation)PSDNorm · LODO, balanced@400, mean over datasetsbalanced accuracy: 77.79 -> 78.34Improved
use PSDNorm instead of BatchNorm in CNNTransformerCNNTransformer · LODO, balanced@400, mean over datasetsbalanced accuracy: 76.38 -> 77.83Improved
use TMA preprocessing in CNNTransformer instead of BatchNormCNNTransformer · LODO, balanced@400, mean over datasetsbalanced accuracy: 76.38 -> 76.30No clear effect
increase PSDNorm filter size f from 5 to very large (e.g., 21) (ablation)PSDNorm · U-Sleep balanced@400balanced accuracyHurt
use PSDNorm filter size f in 5-11 range (ablation)PSDNorm · U-Sleep balanced@400balanced accuracyNo clear effect
replace BatchNorm with PSDNorm (f=5) in training timeU-Sleep · training time per epochseconds per epoch: 161.94 -> 172.85Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy