Decoder only time series models like TimesFM (google), Lag-Llama, Moirai

4 replies
127 views

I am just playing with the analysis of open industrial data (mostly time series IoT sensor data). It is my understanding that encoders play a role in detecting anomalies in the data. Can decoder only models such as Lag-Llama, TimesFM, Moirai detect anomalies or they can only predict? thank you.

Dilini Fernando

Expert ⭐️⭐️⭐️⭐️

Hi @sridiyer,

Thank you for sharing this on Hub. Our team is investigating the matter, and we will keep you informed as soon as we have an update.

V

Vibha Srinivasan

Practitioner ⭐️⭐️⭐️

Apologies for the very delayed response @sridiyer . I haven’t worked too much with time series foundation models, but my understanding is that they can also be used for anomaly detection. The anomaly detection part is done post-prediction, by comparing the forecasted and observed time series, and using techniques for detecting significant differences between their values. The accuracy of course depends on the variety of data that was used for training these models.

Dilini Fernando

Expert ⭐️⭐️⭐️⭐️

Hi @sridiyer,

We are following up to see whether you're satisfied with the responses you've received?

S

sridiyerAuthor

Seasoned ⭐️

I am satisfied with the answer. Trying to carry out one end to end example with (any) time series foundation model for industrial IoT data analysis - kind of stuck, not too many examples online. If anyone has an example (or pointer to an example) will appreciate. thank you very much for checking. @Dilini Fernando

Sign up

Welcome to Cognite Hub