Summary
This paper (arXiv:2607.05393) by Raphaël Bonnet-Guerrini, Bruno Sanchez, Dominique Fouchez, and colleagues presents a deep learning framework for real-bogus classification in time-domain astronomical surveys that requires no human labels. Since reliable annotations are expensive and community-provided labels are noisy and survey-dependent, the authors train on injected transients and bogus-dominated survey data instead. The framework remains robust under strong class contamination and provides calibrated uncertainty quantification. A dual-network model with asymmetric co-teaching handles classes with different label-noise levels, and latent-space visualization is used to interpret learned representations. The authors also propose a low-cost hybrid uncertainty quantification strategy that leverages the dual-network setup to improve calibration. Results show that injection-driven weak supervision enables scalable, consistent real-bogus classification without human labels, offering a practical approach for automated transient discovery pipelines in large surveys.
Paper Overview
Field: AI / Astronomy
Authors: Raphaël Bonnet-Guerrini, Bruno Sanchez, Dominique Fouchez, Benjamin Racine, Maya Guy, Mariam Sabalbal, Manal Yassine, Vincenzo Piuri
Published: 2026-07-06
arXiv: 2607.05393
Abstract
Time-domain surveys produce large numbers of transient candidates, and real-bogus classification is a key step in automated discovery pipelines. Reliable labels are costly, while community labels are often noisy and survey-dependent. This work develops a real-bogus classification framework that requires no human-labeled data: it is trained using injected transients and bogus-dominated survey data, remaining robust under strong class contamination while providing calibrated uncertainty quantification.
Key contributions:
- No human labels required — training relies on injected transients and survey data dominated by bogus detections.
- Dual-network model with asymmetric co-teaching to handle classes with different label-noise levels.
- Latent-space visualization for interpretability of the learned representations.
- Low-cost hybrid uncertainty quantification (UQ) strategy that leverages the dual-network setup to improve calibration.
The results demonstrate that injection-driven weak supervision enables scalable and consistent real-bogus classification without human labels.
---
*Automatically collected on 2026-07-06*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178346215