Train a Sequence Model for Wearable-Telemetry Sleep Staging at a Healthtech
Overview
What this challenge is about.
You receive 220 nights of wearable telemetry from 60 subjects with PSG ground-truth labels. Train three sequence models: an LSTM baseline, a 1D-CNN+GRU hybrid, and a small transformer with sinusoidal positional encoding. Evaluate per-stage macro-F1 and per-stage confusion matrices. Compare to the current rule-based stager. Document on-device feasibility (parameter count, FLOPs per night, model size after int8 quantization).
The Brief
What you'll do, and what you'll demonstrate.
Pick the best sequence architecture for on-device sleep staging by trading off per-stage F1 against parameter count and FLOPs.
Earning criteria — what you'll demonstrate
- Train and compare recurrent and attention-based sequence models
- Reason about on-device deployment constraints early
- Evaluate per-class performance on imbalanced multi-class tasks
- Communicate architecture choices to an engineering audience
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Machine Learning Engineer
Comparing sequence architectures with on-device feasibility analysis is exactly the work MLEs ship at consumer-wearable companies.
This challenge sharpens
- sequence-models
- lstm
- transformers
Applied AI Scientist
Translating PSG-grade evaluation into an engineering recommendation memo is applied-AI-scientist territory in regulated wearable products.
This challenge sharpens
- sequence-models
- evaluation
- on-device-feasibility
MLOps Engineer
On-device feasibility and quantization analysis are increasingly MLOps responsibilities at consumer-device companies.
This challenge sharpens
- on-device-feasibility
- pytorch
- evaluation