Research map / Probabilistic and scientific modeling / Variational and generative sampling
Transformation Autoregressive Networks
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 |
|---|---|---|---|
| Combine flexible transformations with autoregressive conditionals | TAN density estimation · POWER, GAS, HEPMASS, MINIBOONE, BSDS300 | average test log-likelihood: POWER: 0.24 ± 0.01 -> 0.60 ± 0.01; GAS: 10.08 ± 0.02 -> 12.06 ± 0.02; HEPMASS: -15.15 ± 0.02 -> -13.78 ± 0.02; MINIBOONE | Helped |
| Add a linear transformation before LAM + 4xAdd+Re (ablation) | LAM + 4xAdd+Re · Star 128d | held-out test log-likelihood: 77.195 -> 88.837 | Helped |
| Add a linear transformation before LAM + L RNN (ablation) | LAM autoregressive density estimator · Markovian synthetic | held-out test log-likelihood: 14.319 -> 15.658 | Helped |
| Add 4xSRNN+Re and a linear transformation (ablation) | RAM autoregressive density estimator · Markovian synthetic | held-out test log-likelihood: -29.950 -> 16.202 | Helped |
| Add L 4xSRNN+Re transformations (ablation) | LAM autoregressive density estimator · Star 32d | held-out test log-likelihood: -2.041 -> 16.042 | Helped |
| Replace LAM conditionals with independent mixture conditionals (ablation) | L RNN transformation on Markovian synthetic · Markovian synthetic | held-out test log-likelihood: 15.658 -> 5.370 | Hurt |
| Replace RAM conditionals with independent mixture conditionals (ablation) | L RNN+4xAdd+Re transformation · forest | held-out test log-likelihood: 2.672 -> 0.450 | Hurt |
| Remove the initial linear transformation (ablation) | LAM + L 4xAdd+Re · Star 128d | held-out test log-likelihood: 88.837 -> 77.195 | Hurt |
| Choose a single transformation and conditional combination across datasets (ablation) | TAN · all ablation datasets | average performance percentage score: L RNN with RAM: 0.540; L RNN+4xAdd+Re with LAM: 0.336 | Mixed |
| Use TAN density scores for unsupervised anomaly detection | TAN · forest, pendigits, satimage2 | average precision | Helped |
Closest papers by meaning
- Masked Autoregressive Flow for Density Estimation (NeurIPS 2017 · similarity 0.87)
- Autoregressive Energy Machines (ICML 2019 · similarity 0.86)
- Squared Neural Families: A New Class of Tractable Density Models (NeurIPS 2023 · similarity 0.85)
- Nonparametric Estimation of a Factorizable Density using Diffusion Models (NeurIPS 2026 · similarity 0.84)
- Breaking the curse of dimensionality in structured density estimation (NeurIPS 2024 · similarity 0.83)
- Density Ratio Estimation with Conditional Probability Paths (ICML 2025 · similarity 0.82)
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.