Research map / Probabilistic and scientific modeling / Variational and generative sampling
The Convolution Exponential and Generalized Sylvester Flows
Emiel Hoogeboom, Victor Garcia Satorras, Jakub Tomczak, Max Welling
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 |
|---|---|---|---|
| Use a convolution exponential instead of 1 × 1, emerging, or Woodbury mixing | Generative flow with affine coupling layers · CIFAR10 | -ELBO: 3.245 ± 0.002 -> 3.237 ± 0.002 | Improved |
| Use a convolution exponential instead of 1 × 1, emerging, or Woodbury mixing | Generative flow with affine coupling layers · CIFAR10 | NLL: 3.226 ± 0.002 -> 3.218 ± 0.003 | Improved |
| Use Convolutional Sylvester Flows instead of residual block flows | Generative flow · CIFAR10, uniform dequantization | -ELBO: 3.37 -> 3.32 | Improved |
| Use Convolutional Sylvester Flows instead of residual block flows | Generative flow · CIFAR10, variational dequantization | -ELBO: 3.26 -> 3.21 | Improved |
| Add the generalized autoregressive function f_AR (ablation) | Convolutional Sylvester Flows · CIFAR10 | -ELBO: 3.44 -> 3.21 | Improved |
| Add the generalized autoregressive function f_AR (ablation) | Convolutional Sylvester Flows · CIFAR10 | NLL: 3.42 -> 3.19 | Improved |
| Add the basis change (ablation) | Convolutional Sylvester Flows · CIFAR10 | -ELBO: 3.27 -> 3.21 | Improved |
| Add the basis change (ablation) | Convolutional Sylvester Flows · CIFAR10 | NLL: 3.25 -> 3.19 | Improved |
| Add a graph convolution exponential before each coupling layer | Graph Normalizing Flow · MoG-4, MoG-9, MoG-16, MoG-Ring | per-node NLL: 3.69 ± 0.016 -> 3.68 ± 0.017; 4.60 ± 0.067 -> 4.52 ± 0.047; 5.38 ± 0.048 -> 5.26 ± 0.047; 4.22 ± 0.025 -> 4.19 ± 0.036 | Improved |
Closest papers by meaning
- Generative Flows with Matrix Exponential (ICML 2020 · similarity 0.91)
- Woodbury Transformations for Deep Generative Flows (NeurIPS 2020 · similarity 0.89)
- Invertible Convolutional Flow (NeurIPS 2019 · similarity 0.88)
- CDFlow: Building Invertible Layers with Circulant and Diagonal Matrices (NeurIPS 2025 · similarity 0.85)
- Expanding Flow Maps (NeurIPS 2026 · similarity 0.84)
- Beyond Squared Error: Exploring Loss Design for Enhanced Training of Generative Flow Networks (ICLR 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.
