Research map / Probabilistic and scientific modeling / Variational and generative sampling
Neural Spline Flows
Conor Durkan, Artur Bekasov, Iain Murray, George Papamakarios
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 |
|---|---|---|---|
| Replace affine elementwise transforms with rational-quadratic splines in coupling flows | Glow-style coupling flow · GAS | test log likelihood (nats): 12.24 ± 0.03 -> 13.09 ± 0.02 | Improved |
| Replace affine elementwise transforms with rational-quadratic splines in autoregressive flows | Autoregressive flow · HEPMASS | test log likelihood (nats): -17.03 ± 0.02 -> -14.01 ± 0.03 | Improved |
| Replace affine transforms with rational-quadratic splines in coupling flows | Glow-style coupling flow · CIFAR-10 8-bit | BPD: 3.41 -> 3.38 | Improved |
| Replace affine transforms with rational-quadratic splines in coupling flows | Glow-style coupling flow · ImageNet64 5-bit | BPD: 1.81 -> 1.77 | Improved |
| Replace affine transforms with rational-quadratic splines in coupling flows | Glow-style coupling flow · ImageNet64 8-bit | BPD: 3.91 -> 3.82 | Improved |
| Replace affine transforms with rational-quadratic splines in coupling flows | Glow-style coupling flow · CIFAR-10 5-bit | BPD: 1.70 -> 1.70 | No clear effect |
| Use RQ-NSF instead of the standard-normal prior and diagonal-normal posterior | VAE · MNIST | ELBO (nats): -85.61 ± 0.51 -> -82.08 ± 0.46 | Improved |
| Use RQ-NSF instead of the standard-normal prior and diagonal-normal posterior | VAE · EMNIST | ELBO (nats): -125.89 ± 0.41 -> -119.49 ± 0.40 | Improved |
| Use RQ-NSF coupling flows rather than affine coupling flows | VAE · MNIST | ELBO (nats): -82.25 ± 0.46 -> -82.08 ± 0.46 | No clear effect |
| Use RQ-NSF autoregressive flows rather than affine autoregressive flows | VAE · EMNIST | ELBO (nats): -119.85 ± 0.40 -> -119.49 ± 0.40 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Glow — Glow: Generative Flow with Invertible 1x1 Convolutions (NeurIPS 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Normalizing Flows With Multi-Scale Autoregressive Priors (CVPR 2020 · similarity 0.87)
- Building Normalizing Flows with Stochastic Interpolants (ICLR 2023 · similarity 0.87)
- Flow++: Improving Flow-Based Generative Models with Variational Dequantization and Architecture Design (ICML 2019 · similarity 0.87)
- Kernelised Normalising Flows (ICLR 2024 · similarity 0.86)
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.86)
- CDFlow: Building Invertible Layers with Circulant and Diagonal Matrices (NeurIPS 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.
