Conformal calibration and look-elsewhere effect in anomaly detection for new-physics searches
arXiv:2606.13780v1 Announce Type: cross Abstract: Machine-learned anomaly detection is reshaping searches for new physics, but it has outrun the statistics used to interpret it. A raw anomaly score has no calibrated meaning, a model that scans many regions inflates the look-elsewhere effect, and the asymptotic significances the field relies on are blind to the background mismodelling that anomaly detectors are especially prone to. We propose a calibration layer, built on conformal prediction, that turns any anomaly score into a defensible significance with distribution-free, finite-sample guar
The proliferation of machine learning in scientific discovery, particularly in high-stakes fields like particle physics, necessitates more robust statistical interpretation methods for anomaly detection, which this paper addresses.
This research provides a critical framework for validating machine-learned anomaly detection in new-physics searches, ensuring that AI-driven discoveries are statistically sound and interpretable, which is paramount for scientific integrity.
The ability to reliably calibrate anomaly scores and mitigate the look-elsewhere effect means that AI-powered scientific discovery can proceed with greater statistical rigor, potentially accelerating verifiable breakthroughs.
- · Particle physicists
- · Machine learning researchers
- · High-energy physics experiments
- · Uncalibrated AI anomaly detection methods
- · Researchers relying on asymptotic significances alone
Physicists gain a statistically robust method to interpret anomalies found by machine learning models.
Increased confidence in AI-driven new-physics discoveries could lead to faster validation of theoretical models.
The application of conformal prediction in this domain could inspire its adoption in other scientific fields requiring robust anomaly detection.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG