Have automated data analysis pipelines improved quasar identification accuracy significantly?

Yes, automated data analysis pipelines have significantly improved quasar identification accuracy, though the archive materials focus on *gravitational wave* detection rather than quasars — I can answer the quasar part directly, but the archive doesn't cover broader quasar surveys.

Quasar Identification: Recent Advances

Two concrete improvements stand out:

Active learning with spectral data: The DESI survey's QuasarNET pipeline achieved a major efficiency gain by using active learning to retrain its neural network [AG-2025.05-022]. The new version met or exceeded the original classification accuracy in *completeness* (finding real quasars) and *purity* (avoiding false positives) using less than one-tenth of the training data. This matters because it shows that smart labeling strategies — picking the most informative spectra to train on, using a Self-Organizing Map to filter out outliers — can dramatically reduce the human effort needed to improve an automated system.

Light-curve-based classification: The Zwicky Transient Facility catalog identified nearly 5 million quasars using machine learning on how their brightness changes over time, achieving 97% F1 score (a standard measure of accuracy balancing precision and recall) [AG-2025.02-222]. Notably, light curves proved more informative than static imaging from other surveys like Pan-STARRS, suggesting that temporal behavior is key to reliable quasar identification.

The archive materials provided don't address quasar identification beyond these two papers — the rest cover gravitational wave detection pipelines.