DESI QSO Anomaly Detection

DESI-QAD

Quasar Anomaly Detection — ML-driven outlier discovery in 1.6M spectra

DESI-QAD uses unsupervised anomaly detection to identify unusual quasar spectra within DESI Data Release 1. By leveraging autoencoder architectures, we discover statistical outliers that may represent rare physical phenomena, unusual accretion physics, or potentially new classes of astronomical objects.

Scientific Motivation

While the average quasar spectrum is well-understood, the full diversity of these objects remains an active area of research. Outliers can reveal:

Approach

  1. Train autoencoder on majority of spectra to learn "normal" features
  2. Calculate reconstruction error as anomaly score
  3. Rank spectra by anomaly score to identify outliers
  4. Visual inspection and cross-matching for validation