Research map / Probabilistic and scientific modeling / Variational and generative sampling
Pathwise Derivatives Beyond the Reparameterization Trick
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 |
|---|---|---|---|
| Use the OMT gradient instead of the RT gradient | Multivariate Normal with Cholesky covariance · Gaussian Process regression (Mauna Loa CO₂) | ELBO at a given training iteration: 270 vs 500 iterations | Helped |
| Use the OMT gradient instead of the RT gradient | Multivariate Normal with Cholesky covariance · Gaussian Process regression (Mauna Loa CO₂) | wall-clock training efficiency: ∼!1.9 | Mixed |
| Use the OMT gradient instead of the RT gradient | Multivariate Normal · Synthetic test functions, D=50 | gradient variance | Helped |
| Use the OMT gradient instead of the RT gradient | Bivariate Normal · Synthetic test function cos(ω · z), ω=(1,1) | gradient variance | Helped |
| Use the OMT gradient instead of RSVI | Beta distribution · Synthetic test function f(z)=z³, Beta(α, α) | gradient variance | Helped |
| Use the OMT gradient instead of RSVI | Dirichlet distribution · Multinomial-Dirichlet ELBO, 1995 components | gradient variance | Helped |
| Increase RSVI shape augmentation B | RSVI for Dirichlet gradients · Multinomial-Dirichlet ELBO, 1995 components | gradient variance relative to the pathwise gradient | Helped |
| Use the OMT gradient instead of RSVI | Sparse Gamma deep exponential family · Olivetti faces | training ELBO | No clear effect |
| Use the OMT gradient instead of the score function gradient | Univariate mixture of Normals · Synthetic test function f(z)=z⁴ | gradient variance | Helped |
| Use the OMT gradient instead of the Gamma-ratio representation | Beta distribution · Beta(1,1) samples | derivative variance | Helped |
| Use the OMT gradient instead of the RT gradient | Multivariate Normal with Cholesky covariance · Unit Normal with linear test function and κᵢ ∼ 𝒩(0, 1) | total gradient variance: half | Helped |
Closest papers by meaning
- The Generalized Reparameterization Gradient (NeurIPS 2016 · similarity 0.89)
- Approximation Based Variance Reduction for Reparameterization Gradients (NeurIPS 2020 · similarity 0.88)
- Reparameterization Gradient for Non-differentiable Models (NeurIPS 2018 · similarity 0.87)
- Transport meets Variational Inference: Controlled Monte Carlo Diffusions (ICLR 2024 · similarity 0.85)
- On the Relation between Rectified Flows and Optimal Transport (NeurIPS 2025 · similarity 0.85)
- Rao-Blackwellised Reparameterisation Gradients (NeurIPS 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.