Research map / Probabilistic and scientific modeling / Variational and generative sampling
Neural Autoregressive Flows
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 the affine transformer with a DDSF transformer | MAF · POWER | test log-likelihood: 0.48±0.01 -> 0.62 ± 0.01 | Helped |
| replace the affine transformer with a DDSF transformer | MAF · GAS | test log-likelihood: 11.19±0.02 -> 11.91 ± 0.13 | Helped |
| replace the affine transformer with a DDSF transformer | MAF · HEPMASS | test log-likelihood: -15.12±0.02 -> -15.09 ± 0.40 | Helped |
| replace the affine transformer with a DDSF transformer | MAF · MINIBOONE | test log-likelihood: -11.01±0.48 -> -8.86 ± 0.15 | Helped |
| replace the affine transformer with a DDSF transformer | MAF · BSDS300 | test log-likelihood: 157.03±0.07 -> 157.73 ± 0.04 | Helped |
| replace the affine transformer with a DSF transformer | IAF · Binarized MNIST | ELBO: 82.25±0.05 -> 81.92±0.04 | Helped |
| replace the affine transformer with a DSF transformer | IAF · Binarized MNIST | log p(x): 80.05±0.04 -> 79.86±0.01 | Hurt |
| use 10 rather than 5 stacked DDSF transformations (ablation) | MAF-DDSF · POWER | test log-likelihood: 0.62 ± 0.01 -> 0.60 ± 0.02 | Hurt |
| use 10 rather than 5 stacked DDSF transformations (ablation) | MAF-DDSF · GAS | test log-likelihood: 11.91 ± 0.13 -> 11.96 ± 0.33 | Helped |
| use 10 rather than 5 stacked DDSF transformations (ablation) | MAF-DDSF · HEPMASS | test log-likelihood: -15.09 ± 0.40 -> -15.32 ± 0.23 | Hurt |
| use 10 rather than 5 stacked DDSF transformations (ablation) | MAF-DDSF · MINIBOONE | test log-likelihood: -8.86 ± 0.15 -> -9.01 ± 0.01 | Hurt |
| use 10 rather than 5 stacked DDSF transformations (ablation) | MAF-DDSF · BSDS300 | test log-likelihood: 157.73 ± 0.04 -> 157.43 ± 0.30 | Hurt |
Closest papers by meaning
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.87)
- Continuous Autoregressive Modeling with Stochastic Monotonic Alignment for Speech Synthesis (ICLR 2025 · similarity 0.87)
- DELTA-TTS: Adapting Autoregressive Model into a Diffusion Language Model for Text-to-Speech (NeurIPS 2026 · similarity 0.86)
- DiffAR: Denoising Diffusion Autoregressive Model for Raw Speech Waveform Generation (ICLR 2024 · similarity 0.86)
- DiTAR: Diffusion Transformer Autoregressive Modeling for Speech Generation (ICML 2025 · similarity 0.85)
- HAINAN: Fast and Accurate Transducer for Hybrid-Autoregressive ASR (ICLR 2025 · similarity 0.85)
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.