Research map / Probabilistic and scientific modeling / Variational and generative sampling
Masked Autoregressive Flow for 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 |
|---|---|---|---|
| stack autoregressive layers rather than use Real NVP coupling layers | density estimation models · all reported experiments | test log likelihood | Helped |
| increase autoregressive layers from 5 to 10 (ablation) | MAF · GAS | test log likelihood (nats): 9.07 -> 10.08 | Helped |
| increase autoregressive layers from 5 to 10 (ablation) | MAF · HEPMASS | test log likelihood (nats): -17.70 -> -17.73 | Hurt |
| increase autoregressive layers from 5 to 10 (ablation) | MAF · MINIBOONE | test log likelihood (nats): -11.75 -> -12.24 | Hurt |
| increase autoregressive layers from 5 to 10 (ablation) | MAF · CIFAR-10 unconditional | test log likelihood (nats): 2936 -> 3049 | Helped |
| increase autoregressive layers from 5 to 10 (ablation) | MAF · CIFAR-10 conditional | test log likelihood (nats): 2983 -> 3058 | Helped |
| increase autoregressive layers from 5 to 10 (ablation) | MAF · POWER | test log likelihood (nats): 0.14 -> 0.24 | Helped |
| replace a standard Gaussian base with a MADE MoG base (ablation) | MAF (5) · BSDS300 | test log likelihood (nats): 155.69 -> 156.36 | Helped |
| replace a standard Gaussian base with a MADE MoG base (ablation) | MAF (5) · HEPMASS | test log likelihood (nats): -17.70 -> -17.39 | Helped |
| replace a standard Gaussian base with a MADE MoG base (ablation) | MAF (5) · GAS | test log likelihood (nats): 9.07 -> 9.59 | Helped |
| increase the flexibility of MADE conditionals using a Gaussian mixture (ablation) | MADE · CIFAR-10 unconditional | test log likelihood (nats): 147 -> -397 | Hurt |
| use MAF rather than MADE MoG | general-purpose density estimation · reported datasets | test log likelihood | Mixed |
Closest papers by meaning
- Neural Autoregressive Flows (ICML 2018 · similarity 0.91)
- Transformation Autoregressive Networks (ICML 2018 · similarity 0.87)
- Autoregressive Energy Machines (ICML 2019 · similarity 0.86)
- Improved Variational Inference with Inverse Autoregressive Flow (NeurIPS 2016 · similarity 0.86)
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.86)
- Diffusion Beats Autoregressive in Data-Constrained Settings (NeurIPS 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.