Improving SLDS Performance Using Explicit Duration Variables with Infinite Support

Mikolaj Slupinski, Piotr Lipinski

Abstract: Switching Linear Dynamical Systems (SLDS) are probabilistic graphical models used both for self-supervised segmentation and dimensionality reduction. Despite their modeling capabilities, SLDS are particularly hard to train. They oftentimes over-segment the timeseries or completely ignore some of the states, reducing the usefulness of the acquired segmentation.