Research map / Optimization / Optimizers
Convex Potential Flows: Universal Probability Distributions with Optimal Transport and Convex Optimization
Chin-Wei Huang, Ricky T. Q. Chen, Christos Tsirigotis, Aaron Courville
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 |
|---|---|---|---|
| Use CP-Flow instead of the strongest listed baseline | tabular density estimation · Power | negative log-likelihood (nats): -0.62$\,\pm\,$0.01 -> -0.52$\,\pm\,$0.01 | Hurt |
| Use CP-Flow instead of the strongest listed baseline | tabular density estimation · Gas | negative log-likelihood (nats): -11.96$\,\pm\,$0.33 -> -10.36$\,\pm\,$0.03 | Hurt |
| Use CP-Flow instead of the strongest listed baseline | tabular density estimation · HEPMASS | negative log-likelihood (nats): 14.92$\,\pm\,$0.08 -> 16.93$\,\pm\,$0.08 | Hurt |
| Use CP-Flow instead of the strongest listed baseline | tabular density estimation · MINIBOONE | negative log-likelihood (nats): 8.86$\,\pm\,$0.15 -> 10.58$\,\pm\,$0.07 | Hurt |
| Use CP-Flow instead of the strongest listed baseline | tabular density estimation · BSDS300 | negative log-likelihood (nats): -157.73$\,\pm\,$0.04 -> -154.99$\,\pm\,$0.08 | Hurt |
| Use CP-Flow instead of the strongest listed baseline | image density estimation · MNIST | bits/dim: 0.97 -> 1.02 | Hurt |
| Use CP-Flow instead of the strongest listed baseline | image density estimation · CIFAR-10 | bits/dim: 3.28 -> 3.40 | Hurt |
| Replace CP-Flow blocks with coupling blocks (ablation) | CP-Flow image model · MNIST | bits/dim: 1.02 -> 1.02 | No clear effect |
| Replace CP-Flow blocks with coupling blocks (ablation) | CP-Flow image model · CIFAR-10 | bits/dim: 3.40 -> 3.58 | Hurt |
| Replace CP-Flow blocks with invertible residual blocks (ablation) | CP-Flow image model · MNIST | bits/dim: 1.02 -> 1.04 | Hurt |
| Replace CP-Flow blocks with invertible residual blocks (ablation) | CP-Flow image model · CIFAR-10 | bits/dim: 3.40 -> 3.46 | Hurt |
| Add input-augmented ICNN architecture (ablation) | CP-Flow variational inference · Freyfaces | negative ELBO: 4.47 -> 4.45 | Improved |
Closest papers by meaning
- Optimal transport mapping via input convex neural networks (ICML 2020 · similarity 0.88)
- Expectile Regularization for Fast and Accurate Training of Neural Optimal Transport (NeurIPS 2024 · similarity 0.88)
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning (NeurIPS 2025 · similarity 0.86)
- Flow Matching from Viewpoint of Proximal Operators (NeurIPS 2026 · similarity 0.85)
- Functional Gradient Flows for Constrained Sampling (NeurIPS 2024 · similarity 0.85)
- Annealing Flow Generative Models Towards Sampling High-Dimensional and Multi-Modal Distributions (ICML 2025 · 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.
