Research map / Time series / Causal and generative time series
End-to-End Probabilistic Framework for Learning with Hard Constraints
Utkarsh Utkarsh, Danielle C. Maddix, Ruijun Ma, Michael W. Mahoney, Bernie Wang
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 end-to-end with the hard constraint rather than apply it only at inference (ablation) | ProbHardE2E vs ProbConserv · Nonlinear PME m in [4,5] | MSE x1e-6: 276.7 -> 1.59 | Improved |
| use CRPS loss instead of NLL (ablation) | ProbHardE2E-Or · Advection PDE | CRPS x1e-3: 8.669 -> 2.94 | Improved |
| use CRPS loss instead of NLL (ablation) | HardC · PME PDE | CRPS x1e-3: 1.667 -> 1.785 | Hurt |
| use CRPS loss instead of NLL (ablation) | ProbHardE2E and baselines · PDE datasets | MSE | Mixed |
| orthogonal instead of oblique projection on easy and medium PDEs (ablation) | ProbHardE2E · Heat, PME, Advection | CRPS: 10-30% lower CRPS | Improved |
| oblique instead of orthogonal projection on hard Stefan problem (ablation) | ProbHardE2E · Stefan PDE | CRPS x1e-3: 14.147 -> 7.52 | Improved |
| orthogonal instead of oblique projection on hierarchical forecasting (ablation) | ProbHardE2E · Labour | CRPS x1e-3: 36.1 -> 28.6 | Improved |
| orthogonal instead of oblique projection (ablation) | ProbHardE2E · Tourism-L, Traffic, Wiki | CRPS x1e-3: 155.2 -> 156.4 (Tourism-L) | Hurt |
| sampling-free closed-form CRPS with projection on distribution parameters instead of 100 posterior samples (ablation) | ProbHardE2E · Hierarchical forecasting and PDE training time | epoch time: 3.3-4.6x increase with sampling | Improved |
| add DPPL constraint projection to unconstrained backbone (ablation) | VarianceNO · Nonlinear PME | constraint error: 0.020 -> 0 | Improved |
| add convex TVD inequality constraint | ProbHardE2E · PME shock | shock location and negative samples | Improved |
| soft penalty constraints instead of hard (ablation) | SoftC · PDE datasets | constraint error: 0.852 (Heat) vs 0 | Hurt |
Closest papers by meaning
- Probabilistic Forecasting with Stochastic Interpolants and Föllmer Processes (ICML 2024 · similarity 0.86)
- ProbTS: Benchmarking Point and Distributional Forecasting across Diverse Prediction Horizons (NeurIPS 2024 · similarity 0.86)
- Loss Shaping Constraints for Long-Term Time Series Forecasting (ICML 2024 · similarity 0.85)
- $K^2$VAE: A Koopman-Kalman Enhanced Variational AutoEncoder for Probabilistic Time Series Forecasting (ICML 2025 · similarity 0.85)
- Multivariate Probabilistic Time Series Forecasting with Correlated Errors (NeurIPS 2024 · similarity 0.84)
- Uncertainty Quantification for Deep Regression using Contextualised Normalizing Flows (NeurIPS 2025 · similarity 0.84)
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.
