FG-Lasso
A sparse CP decomposition method combining element-wise penalties with fiber-level group Lasso penalties. It can simultaneously detect local anomalies and structured temporal fiber-level anomalies, making it suitable for fault scenarios involving persistent sensor deviations.
Timeline (last 90 days)
FG-Lasso combines element-wise penalties with fiberwise penalties, enabling it to identify structured anomalies concentrated within a specific temporal fiber
Both methods are estimated via alternating optimization, and the sparse components have closed-form solution updates
In nearly all settings of the first simulation study, FG-Lasso achieved or tied for the highest macro F1 score
CP decomposition retains the structural information across the three dimensions of samples, sensors, and time, enabling simultaneous determination of whether an anomaly has occurred and localization of the sensor and time period where it occurred
All Facts (4)
FG-Lasso combines element-wise penalties with fiberwise penalties, enabling it to identify structured anomalies concentrated within a specific temporal fiber
50%UnverifiedBoth methods are estimated via alternating optimization, and the sparse components have closed-form solution updates
50%UnverifiedIn nearly all settings of the first simulation study, FG-Lasso achieved or tied for the highest macro F1 score
50%UnverifiedCP decomposition retains the structural information across the three dimensions of samples, sensors, and time, enabling simultaneous determination of whether an anomaly has occurred and localization of the sensor and time period where it occurred
50%