Research map / Image generation / Fast diffusion sampling
Autoregressive Quantile Networks for Generative Modeling
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 |
|---|---|---|---|
| Replace PixelCNN's softmax likelihood and KL loss with autoregressive quantile regression | Gated PixelCNN on CIFAR-10 · CIFAR-10 | Inception score: 4.6 -> 5.3 | Helped |
| Replace PixelCNN's softmax likelihood and KL loss with autoregressive quantile regression | Gated PixelCNN on CIFAR-10 · CIFAR-10 | FID: 65.9 -> 49.5 | Helped |
| Replace PixelCNN's softmax likelihood and KL loss with autoregressive quantile regression | Gated PixelCNN (15-layer) · ImageNet 32x32 | Inception score: 5.1 -> 7.3 | Helped |
| Replace PixelCNN's softmax likelihood and KL loss with autoregressive quantile regression | Gated PixelCNN (15-layer) · ImageNet 32x32 | FID: 66.4 -> 38.5 | Helped |
| Add class-label conditioning | PixelIQN · ImageNet 32x32 | Inception score | Helped |
| Add class-label conditioning | PixelIQN · ImageNet 32x32 | FID | Helped |
| Use regular quantile regression loss instead of Huber quantile loss (ablation) | PixelIQN · CIFAR-10 and ImageNet 32x32 | sample quality | Hurt |
| Vary the Huber loss parameter κ (ablation) | PixelIQN · CIFAR-10 and ImageNet 32x32 | sample quality | No clear effect |
| Model the VAE latent-code distribution with AIQN instead of sampling from the VAE prior | VAE on CelebA 64x64 · CelebA 64x64 | perceptual quality | Helped |
Closest papers by meaning
- Understand Before You Generate: Self-Guided Training for Autoregressive Image Generation (NeurIPS 2025 · similarity 0.85)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.85)
- Simulation-Based Inference with Quantile Regression (ICML 2024 · similarity 0.85)
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.85)
- Stabilize the Latent Space for Image Autoregressive Modeling: A Unified Perspective (NeurIPS 2024 · similarity 0.84)
- A Spark of Vision-Language Intelligence: 2-Dimensional Autoregressive Transformer for Efficient Finegrained Image Generation (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.