Problem: unsupervised anomaly detection

for seasonal KPIs in web applications. 

Donut: an unsupervised anomaly detection algorithm based on VAE.

Background:

有的time series data have seasonal patterns occurring at regular intervals. 

Data: KPI shapes: seasonal patterns and local variations, noises.  

"abnormal": anomalies and missing points; detect missing points is straightforward.

Existing methods suffer from: 这里面简直是胡说八道。

1. the hassle麻烦 of algorithm picking
2. parameter tuning
3. heavy reliance on labels
4. unsatisfying performance
5. lack of theoretical foundations

Methodology:

VAE is not a sequential model!!!!!!!!!!!!!!!! thus they apply sliding windows. 

在训练时，the anomalies and missing points in a testing window x can bring bias to the mapped z, and further make the reconstruction probability inaccurate. 

如何避免anomalies and missing points对训练造成的biase:

1. missing points. adopt the MCMC-based missing data imputation technique with the trained VAE. 即模拟出missing points的可能值，然后用可能值，代替missing points 的值。
2. anomalies

All the algorithms evaluated in this paper compute one anomaly score for each point. A threshold can be chosen to do the decision: if the score for a point is greater than the threshold, an alert should be triggered

PP: Unsupervised anomaly detection via variational auto-encoder for seasonal KPIs in web applications