Research map / Probabilistic and scientific modeling / Dynamical systems
Stochastic Lifting for Generating Trajectories of Stochastic Physical Systems
Jules Berman, Tobias Blickhan, Benjamin Peherstorfer
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 |
|---|---|---|---|
| add stochastic lifting (SL, 1 step) in place of ARDM with 40 diffusion steps | autoregressive generation · Wave | W2 crossing time: 1.11e-2 -> 4.53e-3 | Improved |
| add stochastic lifting (1 step) in place of ARDM with 40 steps | autoregressive generation · Flow | W2 crossing time: 1.86e-3 -> 9.47e-4 | Improved |
| add stochastic lifting (1 step) in place of ARDM with 40 steps | autoregressive generation · Flow | W2 integrated mass: 6.00e-4 -> 6.08e-4 | No clear effect |
| increase ARDM diffusion steps from 2 to 20 (ablation) | ARDM · Flow | W2 crossing time: 1.18e-1 -> 3.25e-2 | Improved |
| increase ARDM diffusion steps from 20 to 40 (ablation) | ARDM · Wave | W2 crossing time: 7.28e-3 -> 1.11e-2 | Hurt |
| increase ARDM diffusion steps from 2 to 40 (ablation) | ARDM · Flow | W2 integrated mass: 3.46e-2 -> 6.00e-4 | Improved |
| shuffle a growing share of training pairs (x_t, x_t+1) (ablation) | Stochastic Lifting · Duffing oscillator (SDE) | W2 error | Hurt |
| increase label dimension until interpolation is reached (ablation) | Stochastic Lifting · Duffing oscillator (SDE) | final training loss: one order-of-magnitude lower L2 final training loss | Improved |
| increase label dimension (ablation) | Stochastic Lifting · Duffing oscillator (SDE) | network gradient norm | Improved |
| increase label dimension beyond the interpolation threshold (ablation) | Stochastic Lifting · Wave | W2 crossing time | No clear effect |
| map directly from initial condition to final state instead of sequential rollout (ablation) | Stochastic Lifting · Wave and Flow | sample quality | Hurt |
| add stochastic lifting versus the best one-step baseline NUWA | one-step video generation · BAIR video generation | FVD: 86.9 -> 69.0 | Improved |
Closest papers by meaning
- First-Order Trajectory Matching: Fast Ensemble Predictions of Chaotic, Turbulent, Stochastic Systems (NeurIPS 2026 · similarity 0.84)
- Probabilistic Forecasting with Stochastic Interpolants and Föllmer Processes (ICML 2024 · similarity 0.84)
- Itô maps for any-step SDEs (NeurIPS 2026 · similarity 0.84)
- Stochastic Interpolants with Data-Dependent Couplings (ICML 2024 · similarity 0.84)
- Partially Observed Trajectory Inference using Optimal Transport and a Dynamics Prior (ICLR 2025 · similarity 0.83)
- Doob's Lagrangian: A Sample-Efficient Variational Approach to Transition Path Sampling (NeurIPS 2024 · similarity 0.83)
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.
