Research map / Image generation / Fast diffusion sampling
Parallel Multiscale Autoregressive Density Estimation
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 multiscale PixelCNN instead of PixelCNN | conditional image generation · CUB | negative conditional log-likelihood (nats per sub-pixel), validation: 2.93 -> 2.99 | Hurt |
| Use multiscale PixelCNN instead of PixelCNN | conditional image generation · MPII | negative conditional log-likelihood (nats per sub-pixel), validation: 2.92 -> 3.03 | Hurt |
| Use multiscale PixelCNN instead of PixelCNN | conditional image generation · MS-COCO | negative conditional log-likelihood (nats per sub-pixel), validation: 3.08 -> 3.16 | Hurt |
| Use O(T) VPN instead of the O(T) baseline | video generation · Robot Pushing validation | negative log-likelihood (nats per sub-pixel): 2.06 -> 1.03 | Helped |
| Add spatial dependencies with O(T log N) VPN instead of O(T) VPN | video generation · Robot Pushing validation | negative log-likelihood (nats per sub-pixel): 1.03 -> 0.74 | Helped |
| Add spatial dependencies with O(T log N) VPN instead of O(T) VPN | video generation · Robot Pushing test seen | negative log-likelihood (nats per sub-pixel): 1.04 -> 1.06 | Hurt |
| Add spatial dependencies with O(T log N) VPN instead of O(T) VPN | video generation · Robot Pushing test novel | negative log-likelihood (nats per sub-pixel): 1.04 -> 0.97 | Helped |
| Use O(T log N) VPN instead of O(TN) VPN | video generation · Robot Pushing validation | negative log-likelihood (nats per sub-pixel): 0.62 -> 0.74 | Hurt |
| Use the multiscale model instead of pixel-autoregressive density baselines | class-conditional image generation · ImageNet 128 | negative log-likelihood (bits per sub-pixel) | Mixed |
Closest papers by meaning
- Few-shot Autoregressive Density Estimation: Towards Learning to Learn Distributions (ICLR 2018 · similarity 0.89)
- PixelSNAIL: An Improved Autoregressive Generative Model (ICML 2018 · similarity 0.87)
- GENERATING HIGH FIDELITY IMAGES WITH SUBSCALE PIXEL NETWORKS AND MULTIDIMENSIONAL UPSCALING (ICLR 2019 · similarity 0.86)
- Rethinking Generative Image Pretraining: How Far Are We From Scaling Up Next-Pixel Prediction? (ICML 2026 · similarity 0.86)
- L2P: Unlocking Latent Potential for Pixel Generation (NeurIPS 2026 · similarity 0.84)
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 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.