Research map / Probabilistic and scientific modeling / Variational and generative sampling
Sliced Iterative Normalizing Flows
Biwei Dai, Uros Seljak
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 SIG instead of SWF for image generation | Iterative image generators · MNIST | FID: 225.1 -> 4.5 | Improved |
| Use SIG instead of SWF for image generation | Iterative image generators · Fashion-MNIST | FID: 207.6 -> 13.7 | Improved |
| Use SIG instead of the best listed GAN baseline | Image generators · CIFAR-10 | FID: 55.2 -> 66.5 | Hurt |
| Use SIG instead of the best listed GAN baseline | Image generators · CelebA | FID: 30.0 -> 37.3 | Hurt |
| Use SIG rather than GIS for Fashion-MNIST-trained OoD detection | SINF · MNIST OoD | AUROC: 0.824 -> 0.980 | Improved |
| Use SIG rather than GIS for Fashion-MNIST-trained OoD detection | SINF · OMNIGLOT OoD | AUROC: 0.891 -> 0.993 | Improved |
| Apply SIG post-processing to pretrained generator samples | SWF on Fashion-MNIST · Fashion-MNIST | FID: 207.6 -> 23.9 | Improved |
| Apply SIG post-processing to pretrained generator samples | Flow-GAN (ADV) on Fashion-MNIST · Fashion-MNIST | FID: 216.9 -> 21.2 | Improved |
| Apply SIG post-processing to pretrained generator samples | MAF on Fashion-MNIST · Fashion-MNIST | FID: 81.2 -> 16.6 | Improved |
| Choose optimized max K-SWD axes instead of random axes (ablation) | SIG · MNIST | marginal transformations to converge: two orders of magnitude fewer marginal transformations | Improved |
| Use optimized axes rather than random axes (ablation) | SIG · MNIST samples | sample quality | Improved |
| Add hierarchical patch-based modeling after whole-image modeling (ablation) | SIG · MNIST samples | sample quality | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- FFJORD — FFJORD: Free-Form Continuous Dynamics for Scalable Reversible Generative Models (ICLR 2019) · better on 0 of 2 shared results
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 Iterative Denoising (ICML 2026 · similarity 0.86)
- Gradient Boosted Normalizing Flows (NeurIPS 2020 · similarity 0.86)
- Densely connected normalizing flows (NeurIPS 2021 · similarity 0.85)
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.85)
- STARFlow: Scaling Latent Normalizing Flows for High-resolution Image Synthesis (NeurIPS 2025 · similarity 0.84)
- CDFlow: Building Invertible Layers with Circulant and Diagonal Matrices (NeurIPS 2025 · 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.
