Research map / Probabilistic and scientific modeling / Variational and generative sampling
Sticking the Landing: Simple, Lower-Variance Gradient Estimators for Variational Inference
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 |
|---|---|---|---|
| Remove the score-function gradient term | VAE · MNIST, 1 stochastic layer, k=1 | NLL: 86.76 -> 86.40 | Helped |
| Remove the score-function gradient term | VAE · MNIST, 1 stochastic layer, k=50 | NLL: 86.35 -> 86.48 | Hurt |
| Remove the score-function gradient term | IWAE · MNIST, 1 stochastic layer, k=50 | NLL: 84.78 -> 84.45 | Helped |
| Remove the score-function gradient term | VAE · MNIST, 2 stochastic layers, k=1 | NLL: 85.33 -> 84.77 | Helped |
| Remove the score-function gradient term | IWAE · MNIST, 2 stochastic layers, k=50 | NLL: 82.90 -> 83.16 | Hurt |
| Remove the score-function gradient term | VAE · Omniglot, 1 stochastic layer, k=5 | NLL: 107.62 -> 107.40 | Helped |
| Remove the score-function gradient term | IWAE · Omniglot, 1 stochastic layer, k=50 | NLL: 104.67 -> 104.16 | Helped |
| Remove the score-function gradient term | VAE · Omniglot, 2 stochastic layers, k=1 | NLL: 107.58 -> 105.22 | Helped |
| Remove the score-function gradient term | IWAE · Omniglot, 2 stochastic layers | NLL: The paper reports improvements of 2.34, 1.2, and 0.52 nats for k={1, 5, 50}. | Helped |
| Use the path derivative estimator rather than the total derivative estimator | VAE and IWAE · MNIST and Omniglot, reported experiments | NLL | Mixed |
Closest papers by meaning
- VarGrad: A Low-Variance Gradient Estimator for Variational Inference (NeurIPS 2020 · similarity 0.93)
- Marginalized Stochastic Natural Gradients for Black-Box Variational Inference (ICML 2021 · similarity 0.92)
- Variational Inference for Sequential Data with Future Likelihood Estimates (ICML 2020 · similarity 0.91)
- Variational Inference and Model Selection with Generalized Evidence Bounds (ICML 2018 · similarity 0.91)
- Variational Inference with Mixtures of Isotropic Gaussians (NeurIPS 2025 · similarity 0.90)
- Least squares variational inference (NeurIPS 2025 · similarity 0.90)
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.