Research map / Probabilistic and scientific modeling / Dynamical systems
Learning to Filter with Predictive State Inference Machines
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace back-propagation (random init) with DAgger for training PSIM linear filter (ablation) | PSIM-Linear · Robot Drill, Motion Capture, Beach Video | 1-step filter error: 2.54→2.15, 9.94→5.75, 268.73→164.23 | Helped |
| Use RFF random features instead of linear regression within PSIM-DAgger (ablation) | PSIM with DAgger · Robot Drill, Motion Capture, Beach, Flag video | 1-step filter error: 2.15→1.80, 5.75→5.41, 164.23→130.53, 1.28e3→1.24e3 | Helped |
| Use PSIM with DAgger (linear or RFF) instead of spectral baselines N4SID and IVR | PSIM with DAgger · Robot Drill, Motion Capture, Beach, Flag video | 1-step filter error: outperforms N4SID and IVR on all four datasets | Helped |
| Back-propagation refinement after DAgger initialization (ablation) | PSIM-Linear · Robot Drill, Motion Capture, Beach Video | 1-step filter error: 2.15→2.09, 5.75→5.66, 164.23→164.08 (marginal) | Helped |
| Use k=5 future observations in predictive state instead of k=1 (ablation) | PSIM with DAgger · Robot Drill, Motion Capture, Flag, Beach Video | 1-step filter error: 5% robot assembly, 6% motion capture, 8% flag, 32% beach video | Helped |
| Use RFF-based back-propagation instead of linear back-propagation (ablation) | PSIM back-propagation · Robot Drill, Motion Capture, Beach Video | 1-step filter error: 2.54→2.54 (RFF Bp), beach 268.73→202.10, motion 9.94→9.26 | Helped |
| Use PSIM-RFF with DAgger instead of a recurrent neural network | RNN comparison · Robot Drill, Motion Capture, Beach Video | 1-step filter error: 1.99→1.80, 9.6→5.41, 346.0→130.53 | Helped |
| Use predictive-state features with second moments (phi_2) instead of phi_1 with linear DAgger (ablation) | PSIM-Linear with DAgger · Robot Drill, Motion Capture, Beach, Flag video | 1-step filter error: 2.15→2.05, 5.75→5.47, 164.23→154.02, 1.28e3→1.27e3 | Mixed |
| Forward Training instead of DAgger on the synthetic LDS (ablation) | PSIM with forward training · Synthetic LDS | filter error convergence vs training trajectories | Mixed |
Closest papers by meaning
- Gated Inference Network: Inference and Learning State-Space Models (NeurIPS 2024 · similarity 0.88)
- One Filters All: A Generalist Filter For State Estimation (NeurIPS 2025 · similarity 0.88)
- Markovian Dynamics Enforcer: Feasibility Preserving Correction on Learned Dyanmics Manifolds (NeurIPS 2026 · similarity 0.86)
- Identifying latent state transitions in non-linear dynamical systems (ICLR 2025 · similarity 0.86)
- Environment Inference for Learning Generalizable Dynamical System (NeurIPS 2025 · similarity 0.86)
- UnHiPPO: Uncertainty-aware Initialization for State Space Models (ICML 2025 · similarity 0.86)
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.