Prototype a Normalizing Flow for Anomaly Scoring in Climate Sensor Data
Overview
What this challenge is about.
You receive 12 months of multivariate sensor traces (8 channels per sensor, hourly). Train a Normalizing Flow (Real NVP or a small Neural Spline Flow) on a clean training window per sensor. Use the flow's log-density as the anomaly score and compare to the current Z-score baseline on a hand-labeled anomaly set (around 300 anomalies). Report ROC-AUC, calibration, and per-anomaly-type sensitivity. Write a 4-page research note.
The Brief
What you'll do, and what you'll demonstrate.
Show whether a Normalizing Flow produces better-calibrated anomaly scores than the current Z-score detector on hand-labeled geothermal sensor data.
Earning criteria — what you'll demonstrate
- Implement and train normalizing flows on multivariate sensor data
- Use density estimates as anomaly scores defensibly
- Evaluate calibration formally (reliability, ECE, risk-coverage)
- Communicate research results to a domain-scientific 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.
ML Researcher
Normalizing flows for anomaly scoring is an active research thread; this challenge produces a credible first artifact.
This challenge sharpens
- normalizing-flows
- density-estimation
- anomaly-detection
Research Scientist
Formal calibration analysis on industrial sensor data is the kind of rigor expected from a junior research scientist.
This challenge sharpens
- normalizing-flows
- calibration
- evaluation
Applied AI Scientist
Beating a deployed Z-score detector with a research method is exactly the bridge applied AI scientists build between research and product.
This challenge sharpens
- density-estimation
- anomaly-detection
- calibration