Research map / Probabilistic and scientific modeling / Variational and generative sampling
Telescoping Density-Ratio Estimation
Benjamin Rhodes, Kai Xu, Michael U. Gutmann
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 a single density-ratio estimate with telescoping bridges | Density-ratio estimation · 1d peaked ratio | ratio estimation error: 100 samples; 100,000 samples | Improved |
| Replace a single density-ratio estimate with telescoping bridges | Density-ratio estimation · High-dimensional Gaussian MI | MI estimation accuracy: 20 nats; 80 nats; 320 dimensional variables | Improved |
| Replace single-ratio baselines with TRE | SpatialMultiOmniglot representation learning · SpatialMultiOmniglot | linear evaluation accuracy: 20-60%; ~3%; >97% | Improved |
| Replace NCE single-ratio estimation with TRE using Gaussian noise | MNIST EBM · MNIST | bpd (RAISE): 1.99 -> 1.35 | Improved |
| Replace NCE single-ratio estimation with TRE using Gaussian Copula noise | MNIST EBM · MNIST | bpd (RAISE): 1.48 -> 1.23 | Improved |
| Replace NCE single-ratio estimation with TRE using RQ-NSF noise | MNIST EBM · MNIST | bpd (RAISE): 1.10 -> 1.09 | Improved |
| Generate linear-combination waymarks in x-space instead of z-space (ablation) | TRE with Gaussian Copula noise · MNIST | bpd (RAISE): 1.23 -> 1.33 | Hurt |
| Vary the number of waymarks between 5-30 (ablation) | TRE with Gaussian Copula noise · MNIST | approximate log-likelihood (bpd): no greater than 0.03 bpd | No clear effect |
| Add an attention layer (ablation) | TRE MNIST architecture · MNIST | estimated log-likelihood | No clear effect |
| Add spectral normalisation (ablation) | TRE MNIST architecture · MNIST | performance | No clear effect |
| Add Instance Normalisation (ablation) | TRE energy-based modelling · Energy-based modelling | performance | Hurt |
| Use average pooling rather than max pooling (ablation) | TRE SpatialMultiOmniglot architecture · SpatialMultiOmniglot | final performance | Mixed |
Closest papers by meaning
- Density Ratio Estimation with Conditional Probability Paths (ICML 2025 · similarity 0.89)
- Overcoming Saturation in Density Ratio Estimation by Iterated Regularization (ICML 2024 · similarity 0.87)
- Binary Losses for Density Ratio Estimation (ICLR 2025 · similarity 0.86)
- Classification Diffusion Models: Revitalizing Density Ratio Estimation (NeurIPS 2024 · similarity 0.85)
- Density Ratio Estimation-based Bayesian Optimization with Semi-Supervised Learning (ICML 2025 · similarity 0.84)
- Projection Pursuit Density Ratio Estimation (ICML 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.
