Research map / Architectures and efficiency / CNN design
ButterflyFlow: Building Invertible Layers with Butterfly Matrices
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 invertible butterfly layers in place of the baseline linear transformations | Glow-based flow models · Permuted CIFAR-10 | bits per dimension: 5.48 -> 5.11 | Helped |
| Use invertible butterfly layers in place of the baseline linear transformations | Glow-based flow models · Permuted ImageNet-32×32 | bits per dimension: 6.29 -> 6.18 | Helped |
| Use invertible butterfly layers in place of the baseline linear transformations | Glow-based flow models · Permuted MNIST | bits per dimension: 1.44 -> 1.42 | Helped |
| Use invertible butterfly layers in place of baseline linear transformations | Glow-based flow models · Galaxy images | bits per dimension: 1.98 -> 1.95 | Helped |
| Use invertible butterfly layers in place of baseline linear transformations | Glow-based flow models · MIMIC-III average | negative log-likelihood per dimension: -11.47 -> -27.92 | Helped |
| Use invertible butterfly layers | Glow-based flow models · MNIST | bits per dimension: 1.05 | No clear effect |
| Use invertible butterfly layers | Glow-based flow models · CIFAR-10 | bits per dimension: 3.35 -> 3.33 | No clear effect |
| Use invertible butterfly layers | Glow-based flow models · ImageNet-32×32 | bits per dimension: 4.09 | No clear effect |
| Use ButterflyFlow instead of the second-best-performing baseline | MIMIC-III waveform models · MIMIC-III | number of parameters: 42,576 -> 15,280 | Helped |
Closest papers by meaning
- CDFlow: Building Invertible Layers with Circulant and Diagonal Matrices (NeurIPS 2025 · similarity 0.89)
- Invertible Convolutional Flow (NeurIPS 2019 · similarity 0.87)
- Densely connected normalizing flows (NeurIPS 2021 · similarity 0.87)
- Woodbury Transformations for Deep Generative Flows (NeurIPS 2020 · similarity 0.85)
- Kernelised Normalising Flows (ICLR 2024 · similarity 0.84)
- Normalizing Flows are Capable Generative Models (ICML 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.