[Open DMQA Semiar] Transformer-Based Anomaly Detection in Multivariate Time Series

As a significant portion of data collected in industry takes the form of multivariate time-series data, it is crucial to properly understand and apply it. In particular, the field of outlier detection, which is suitable for detecting defects in situations involving a large amount of normal data, continues to receive steady attention, and research on outlier detection methodologies suitable for multivariate time-series data is actively being conducted. Recently, various deep learning-based methodologies have been proposed, reflecting complex time-series characteristics and relationships between variables, thereby achieving remarkable performance improvements. This seminar aims to examine everything from the definition of multivariate time-series data to major deep learning-based methodologies, and further to Transformer-based methodologies, which are currently gaining prominence as state-of-the-arts. References: [1] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems, 30. [2] Xu, J., Wu, H., Wang, J., & Long, M. (2021, September). Anomaly Transformer: Time Series Anomaly Detection with Association Discrepancy. In International Conference on Learning Representations. [3] Tuli, S., Casale, G., & Jennings, N. R. (2022). TranAD: Deep transformer networks for anomaly detection in multivariate time series data. arXiv preprint arXiv:2201.07284.