Research map / Image generation / Fast diffusion sampling
Normalizing Flows With Multi-Scale Autoregressive Priors
Check 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 |
|---|---|---|---|
| Add a multi-scale autoregressive prior to the split coupling flow (ablation) | SCF flow with MixLogCDF couplings · CIFAR10 | bits/dim: 3.29 -> 3.27 | Helped |
| Add a multi-scale autoregressive prior to the split coupling flow | SCF flow with affine couplings · CIFAR10 | bits/dim: 3.35 vs 3.33 | Helped |
| Increase parameter-prediction network width from 256 to 512 channels (ablation) | mAR-SCF with affine couplings · CIFAR10 | bits/dim: 3.33 -> 3.31 | Helped |
| Increase parameter-prediction network width from 256 to 512 channels (ablation) | mAR-SCF with affine couplings · MNIST | bits/dim: 1.04 -> 1.03 | Helped |
| Use 256 rather than 96 channels (ablation) | mAR-SCF with MixLogCDF couplings · CIFAR10 | bits/dim: 3.27 -> 3.24 | Helped |
| Use affine rather than MixLogCDF couplings (ablation) | mAR-SCF · CIFAR10 | FID: 41.9 -> 41.0 | Helped |
| Use affine rather than MixLogCDF couplings (ablation) | mAR-SCF · CIFAR10 | Inception score: 5.7 -> 5.7 | No clear effect |
| Reduce affine-flow parameter-prediction width from 512 to 256 channels (ablation) | mAR-SCF · CIFAR10 | sampling speed (ms/image): 17 -> 6 | Helped |
| Add the multi-scale autoregressive prior | split coupling flow · MNIST | bits/dim: 1.05 vs 1.04 | Helped |
Closest papers by meaning
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.90)
- STARFlow: Scaling Latent Normalizing Flows for High-resolution Image Synthesis (NeurIPS 2025 · similarity 0.90)
- STARFlow2: Bridging Language Models and Normalizing Flows for Unified Multimodal Generation (NeurIPS 2026 · similarity 0.88)
- Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (ICML 2024 · similarity 0.88)
- Boosting Flow-based Generative Super-Resolution Models via Learned Prior (CVPR 2024 · similarity 0.88)
- CAR-Flow: Condition-Aware Reparameterization Aligns Source and Target for Better Flow Matching (NeurIPS 2025 · similarity 0.87)
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.