Closed hojie11 closed 1 month ago
of course, that's a natural transition. Short answer: not yet. Long answer: when we just started working on TSAD, we noticed that there had been a lot of progress illusion due to wrong metrics, as described in our paper. Therefore, we first wanted to build a robust TSAD model for univariate time series, and then move on to multivariate time series because we believe in incremental improvement. Despite the availability of several obvious ways to modify TimeVQVAE-AD for multivariate time series, I don't think our team will be working on it in the near future because of the end of funding project and my PhD program.
But I'm sure it'd be a good research topic with highly promising feasibility.
I also used to think that this method could be applied to AD for MTS. I was trying to do unsupervised anomaly detection for MTS with this method. It's kind of challenging 🤣🤣 BTW, I really appreciate you work. 👍
Nice to hear that! Thanks 👍
Thanks for your nice work. It is really helpful to do anomaly detection on univariate time series. I think your method could be applied to multivariate time series data. have you tried on the multivariate?