Time Series Source Separation with Slow Flows
This work addresses the challenge of identifiable source separation in time series analysis, which is incremental as it builds upon existing methods in blind source separation and flow-based models.
The paper tackled the problem of making time series decomposition identifiable by fitting slow feature analysis into the flow-based models framework, resulting in a method that enables blind source separation for time series.
In this paper, we show that slow feature analysis (SFA), a common time series decomposition method, naturally fits into the flow-based models (FBM) framework, a type of invertible neural latent variable models. Building upon recent advances on blind source separation, we show that such a fit makes the time series decomposition identifiable.