Research map / Probabilistic and scientific modeling / Dynamical systems
Latent Matters: Learning Deep State-Space Models
Alexej Klushyn, Richard Kurle, Maximilian Soelch, Botond Cseke, Patrick van der Smagt
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train with constrained optimisation (CO) instead of annealing (ablation) | EKVAE · Pendulum | MSE (smoothed): 1.94E-3 -> 3.53E-4 | Improved |
| Train with CO instead of annealing (ablation) | DKS · Pendulum | R2 velocity: 0.040 -> 0.962 | Improved |
| Train with CO instead of annealing (ablation) | DVBS · Pendulum | MSE (smoothed): 4.67E-3 -> 9.41E-4 | Improved |
| Add VHP prior to EKVAE (CO) (ablation) | EKVAE · Pendulum | MSE (smoothed): 3.53E-4 -> 1.99E-4 | Improved |
| Add VHP prior to DVBS (CO) (ablation) | DVBS · Pendulum | MSE (smoothed): 9.41E-4 -> 5.63E-4 | Improved |
| Add VHP prior to DKS (CO) (ablation) | DKS · Pendulum | MSE (smoothed): 2.36E-3 -> 1.73E-3 | Improved |
| Higher test ELBO (ablation) | annealing vs CO · Pendulum | test ELBO: 804.0 -> 804.7 | No clear effect |
| Train with CO instead of annealing (ablation) | EKVAE · Reacher angle | MSE (smoothed): 4.38E-4 -> 2.13E-5 | Improved |
| Train with CO instead of annealing (ablation) | DKS · Reacher angle | MSE (smoothed): 1.12E-3 -> 3.52E-4 | Improved |
| Train with CO instead of annealing (ablation) | EKVAE · Reacher RGB | MSE (smoothed): 1.82E-3 -> 1.64E-4 | Improved |
| Use CO with VHP for RNN-based KVAE (ablation) | KVAE · Pendulum | R2 velocity: 0.134 -> 0.043 | No clear effect |
| Use CO with VHP for RNN-based RSSM (ablation) | RSSM · Pendulum | MSE (smoothed): 3.23E-3 -> 3.10E-3 | No clear effect |
Closest papers by meaning
- Efficient Learning of Deep State Space Models via Importance Smoothing (ICML 2026 · similarity 0.88)
- Deep Variational Bayes Filters: Unsupervised Learning of State Space Models from Raw Data (ICLR 2017 · similarity 0.87)
- Gated Inference Network: Inference and Learning State-Space Models (NeurIPS 2024 · similarity 0.86)
- UnHiPPO: Uncertainty-aware Initialization for State Space Models (ICML 2025 · similarity 0.86)
- Identifying latent state transitions in non-linear dynamical systems (ICLR 2025 · similarity 0.85)
- eXponential FAmily Dynamical Systems (XFADS): Large-scale nonlinear Gaussian state-space modeling (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
