Research map / Probabilistic and scientific modeling / Variational and generative sampling
Flowification: Everything is a normalizing flow
Bálint Máté, Samuel Klein, Tobias Golling, François Fleuret
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 |
|---|---|---|---|
| sample augmentation noise from normal instead of uniform (ablation) | fConv1 · CIFAR-10 | BPD: 6.24 -> 4.91 | Improved |
| sample augmentation noise from normal instead of uniform (ablation) | fConv1 · MNIST | BPD: 3.41 -> 3.11 | Improved |
| use non-overlapping kernels (fConv2) instead of overlapping (fConv1) | flowified convolutional network · MNIST | BPD: 3.11 -> 1.41 | Improved |
| use non-overlapping kernels (fConv2) instead of overlapping (fConv1) | flowified convolutional network · CIFAR-10 | BPD: 4.91 -> 4.20 | Improved |
| use convolutional layers instead of linear-only layers | flowified network · MNIST | BPD: 4.19 -> 3.11 | Improved |
| use convolutional layers instead of linear-only layers | flowified network · CIFAR-10 | BPD: 5.45 -> 4.91 | Improved |
| add rational quadratic spline (NSF) layers between flowified layers | fConv1 · MNIST | BPD: 3.11 -> 2.70 | Improved |
| add rational quadratic spline (NSF) layers between flowified layers | fConv1 · CIFAR-10 | BPD: 4.91 -> 3.93 | Improved |
| add rational quadratic spline (NSF) layers between flowified layers | fConv2 · MNIST | BPD: 1.41 -> 1.35 | Improved |
| add rational quadratic spline (NSF) layers between flowified layers | fConv2 · CIFAR-10 | BPD: 4.20 -> 3.69 | Improved |
| use flowified MLP instead of Neural Spline Flow | tabular density estimation · POWER | test log likelihood: 0.63 -> -0.50 | Hurt |
| use flowified MLP instead of Neural Spline Flow | tabular density estimation · GAS | test log likelihood: 13.02 -> 5.35 | Hurt |
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.
- Glow — Glow: Generative Flow with Invertible 1x1 Convolutions (NeurIPS 2018) · better on 0 of 7 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
- Invertible Convolutional Flow (NeurIPS 2019 · similarity 0.88)
- Bidirectional Normalizing Flow: From Data to Noise and Back (CVPR 2026 · similarity 0.88)
- CDFlow: Building Invertible Layers with Circulant and Diagonal Matrices (NeurIPS 2025 · similarity 0.86)
- On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticity (NeurIPS 2025 · similarity 0.85)
- Understanding Likelihood of Normalizing Flow and Image Complexity through the Lens of Out-of-Distribution Detection (AAAI 2024 · similarity 0.85)
- Normalizing Flows are Capable Generative Models (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.
