AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-19
Missed Detections and False Alarms in a CNN Technosignature Detector on Breakthrough Listen Data
Radio technosignature searches use machine-learning detectors to narrow down candidates, but these are usually validated only on data similar to their training set. The author froze a standard competition detector and tested it with 15,000 injected signals across ten morphologies and five amplitudes, placed into real Green Bank backgrounds. Recall did not follow the expected ordering and fell as signals grew brighter, even for needle-like signals the detector was trained on (0.590 to 0.090), and one interference type reached a false positive rate of 0.627.