houjingyi-ustb / discover_PLF

17 stars 3 forks source link

Decomposition of Time Series #1

Open sujing863 opened 1 year ago

sujing863 commented 1 year ago

In your paper, you use the decomposition method, but I don't see the module in the code. Can you tell me where it is?

houjingyi-ustb commented 1 year ago

It is implemented by the 1D convolution operations.

sujing863 commented 1 year ago

In the time series decomposition module of your paper it is written that our method generalizes the data processing from the frequency domain to the concept domain, I would like to know how to push the data to the concept domain using 1D convolution. As well, can you explain the meaning of concept domain?

houjingyi-ustb commented 1 year ago

It means the domain spanned by the latent factors inferred using our method. It is raised in the literature of something like VAEs and disentanglement learning. Maybe you can refer to some related papers, such as "Variational inference of disentangled latent concepts from unlabeled observations" - ICLR2018, for more details.

sujing863 commented 1 year ago

Thanks for your answer.