Use case · Idea to test
EEG sleep-stage labels
Classify recorded sleep segments directly.
- Recorded signal segment
- Predict a sleep stage
- Review the sequence
A sleep-research pipeline could assign a defined stage to each recorded segment, then show the sequence for review. A language model does not need to write the stage name token by token.
Where the small model fits
A signal encoder and classification head return stage scores. Earlier heads or a small-model fallback could be compared while retaining the temporal context needed by the task.
What would need to work
Compare with a conventional sleep-stage classifier at matched quality. Split by participant and recording, measure mistakes between stages and total time. This is not an insomnia diagnosis or a claim that monitoring can safely omit parts of a recording.
Research proposal only. SkinDeep has no clinical model, patient-use demo or validated medical performance for this task.
A dataset to start with
Sleep-EDF Expanded includes sleep recordings and corresponding sleep-stage annotations for research. Sleep-EDF Expanded dataset.