Disentangled State Space Representations
This addresses the challenge of knowledge transfer and robust prediction in sequential data across diverse domains, though it appears incremental as it builds on existing state space and VAE frameworks.
The paper tackles the problem of learning cross-domain sequence representations by introducing disentangled state space models (DSSM) that separate domain-invariant dynamics from domain-specific information, achieving competitive performance in tasks like ODE system identification and bouncing ball video prediction across varying conditions.
Sequential data often originates from diverse domains across which statistical regularities and domain specifics exist. To specifically learn cross-domain sequence representations, we introduce disentangled state space models (DSSM) -- a class of SSM in which domain-invariant state dynamics is explicitly disentangled from domain-specific information governing that dynamics. We analyze how such separation can improve knowledge transfer to new domains, and enable robust prediction, sequence manipulation and domain characterization. We furthermore propose an unsupervised VAE-based training procedure to implement DSSM in form of Bayesian filters. In our experiments, we applied VAE-DSSM framework to achieve competitive performance in online ODE system identification and regression across experimental settings, and controlled generation and prediction of bouncing ball video sequences across varying gravitational influences.