Skip to content

Latest commit

 

History

History
22 lines (13 loc) · 901 Bytes

README.md

File metadata and controls

22 lines (13 loc) · 901 Bytes

LSTM Based Anomaly Detection

Hypothetically it should be possible to autoencoder an LSTM in order to build an anomaly detector. This proves that this is possible.

Example

It is shown that a simple sin wave with added amplitudal noise can be detected if trained with a pure unsupervised LSTM autoencoder.

The entire signal as it is presented to the network [sliding window] :

Alt text

The signal approximation (mean) :

Alt text

The MSE output of the LSTM detector:

Alt text

The number of noise samples added (and caught):

Alt text