Exploiting Variable Correlation with Masked Modeling for Anomaly Detection in Time Series
Online anomaly detection in multi-variate time series is a challenging problem particularly when there is no supervision information. Autoregressive predictive models are often used for this task, but such detection methods often overlook correlations between variables observed the most recent step and thus miss some anomalies that violate normal variable relations. In this work, we propose a masked modeling approach that captures variable relations and temporal relations in a single predictive model. This new method can be combined with a wide range of predictive models. Our empirical study shows that this new modeling method clearly improves detection performance over pure autoregressive models when the time series itself is not very predictable. We also show that detection performance is strongly correlated with validation loss in the new modeling method and thus hyperparameter tuning is still feasible for training anomaly detectors.
Please join meeting in Halligan 209, or via Zoom.
Join Zoom meeting: https://tufts.zoom.us/j/7972618065
Password: see colloqium email
Dial-in option not available for this talk.