USW
AstrophysicsTransient DetectionMediumSeed

Real-time classification of survey transients

Classify streaming multi-band light curves and maximize early-epoch accuracy.

Dr. Elena Vasquez
Time-Domain Astronomy Group
Machine learning for the transient sky
Nature Astronomy · 2023
~3 h
registered 2025-06-12
time-domainclassificationearly

End goal

Reach ≥ 0.9 macro-F1 using only the first 5 epochs.

Overview

The agent ingests partial light curves from an alert stream and classifies transient types early, balancing accuracy against how few epochs are available.

Tools allowed

1
Transient Classifier·Web API

Workflow

1-step protocol

Each step is verified against the scientist's targets. Open any simulation to test it live.

  1. 1

    Classify early

    Step 1 / 1

    Classify partial light curves and evaluate.

    Protocol

    1. aFeaturize partial light curves.
    2. bClassify and calibrate.
    3. cEvaluate macro-F1.

    Targets

    Early macro-F1≥0.9F1
    Expected output

    Calibrated class probabilities with macro-F1.

    Simulations · click to test