| KEYWORD |
Unraveling Anomaly Detection for Multivariate Time Series
keywords ANOMALY DETECTION, DEEP LEARNING, MACHINE LEARNING, DEEP LEARNING, OPTIMIZATION, TIME SERIES
Reference persons SANTA DI CATALDO
External reference persons francesco.ponzio@polito.it, alessio.mascolini@polito.it
Research Groups DAUIN - GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA
Description Research in the development of novel methodologies for detecting anomalies in time series is of great impact and has attracted a huge body of research in the last years. Nonetheless, progress is hampered for the following three main reasons: (1) public benchmarks are biased (e.g., due to potentially erroneous anomaly labels), (2) there is no widely accepted standard evaluation metric, and (3) evaluation protocols are mostly inconsistent.
Starting from the longitudinal study by Wagner et al.[1], the current thesis aims at 1) evaluating the current state of the art panorama for time series anomaly detection (AD) on a new dataset recently published; 2) evaluating the current state of the art panorama for time series anomaly detection (AD) on a real world industrial case study (namely, Electron beam melting); 3) exploring on such datasets a new paradigm for anomaly detection based on variational inference.
[1] Wagner, Dennis, et al. "TimeSeAD: Benchmarking Deep Multivariate Time-Series Anomaly Detection." Transactions on Machine Learning Research (2023).
Required skills Solid programming skills (Python), prior knowledge of Machine Learning/Deep Learning design frameworks.
Deadline 28/07/2024
PROPONI LA TUA CANDIDATURA