RoseCDL: Robust and Scalable Convolutional Dictionary Learning for Rare-event Detection

Jad Yehya, Mansour Benbakoura, Cédric Allain, Benoît Malezieux, Matthieu Kowalski, Thomas Moreau · 2025

Read RoseCDL: Robust and Scalable Convolutional Dictionary Learning for Rare-event Detection by Jad Yehya, Mansour Benbakoura, Cédric Allain, Benoît Malezieux, Matthieu Kowalski, Thomas Moreau (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-07523.txt.

Description

Jad Yehya and 5 co-authors present RoseCDL in 2025, an algorithm for spotting rare events and anomalies in long signals. Many detection problems split into learning common local patterns and flagging deviations from them, in fields from astronomy to biomedicine. Convolutional dictionary learning models local patterns well, but it is computationally heavy and sensitive to outliers, which has limited its use for anomaly hunting.

RoseCDL trains efficiently with stochastic windowing, sampling small chunks of the signal, and detects outliers inline during training to stay robust. Anomalies are then flagged in an unsupervised way from local reconstruction loss: patterns the dictionary cannot rebuild look suspicious. Experiments on real-world datasets show better detection accuracy and computational efficiency than earlier approaches. The authors aim to make dictionary learning practical for large-scale signal analysis.

Catalog record

CREATORJad Yehya, Mansour Benbakoura, Cédric Allain, Benoît Malezieux, Matthieu Kowalski, Thomas Moreau
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE49 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-07523

Browse the full library · About the Tower of Babel