Research map / Optimization / Optimizers
Generalizing Stochastic Smoothing for Differentiation and Gradient Estimation
Felix Petersen, Christian Borgelt, Aashwin Ananda Mishra, Stefano Ermon
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 |
|---|---|---|---|
| use Cauchy distribution with 32,768 samples instead of 256 samples (best cv) (ablation) | generalized stochastic smoothing · 4-digit MNIST sorting (n=5) | exact match accuracy: 55.6 -> 85.1 | Improved |
| use best sampling strategy chosen by cross-validation instead of vanilla MC (Gaussian, 256 samples) (ablation) | generalized stochastic smoothing · 4-digit MNIST sorting (n=5), 256 samples | exact match accuracy: 82.3 -> 83.1 | Improved |
| use best sampling strategy by cv instead of vanilla for Gumbel at 1k samples (ablation) | generalized stochastic smoothing · 4-digit MNIST sorting (n=5), 1k samples | exact match accuracy: 82.0 -> 84.2 | Improved |
| use best sampling strategy by cv instead of vanilla for Cauchy at 256 samples (ablation) | generalized stochastic smoothing · 4-digit MNIST sorting (n=5), 256 samples | exact match accuracy: 68.1 -> 55.6 | Hurt |
| add antithetic sampling on top of QMC/RQMC strategies (ablation) | generalized stochastic smoothing · gradient variance (sorting, shortest paths) | L2 error of gradient | Hurt |
| add antithetic sampling to vanilla MC without LOO covariate (ablation) | vanilla Monte Carlo · gradient variance (sorting, shortest paths) | L2 error of gradient | Improved |
| use Cartesian RQMC instead of MC (ablation) | generalized stochastic smoothing · gradient variance (sorting, shortest paths) | L2 error of gradient | Improved |
| add leave-one-out covariate instead of none (ablation) | generalized stochastic smoothing · gradient variance (sorting, shortest paths) | L2 error of gradient | Improved |
| use f(x) covariate instead of none or LOO (ablation) | generalized stochastic smoothing · 4-digit MNIST sorting (n=5) | exact match accuracy | No clear effect |
| smooth the algorithm instead of the loss (ablation) | generalized stochastic smoothing · Warcraft shortest-path | exact match accuracy | Improved |
| smooth the algorithm instead of the loss (ablation) | generalized stochastic smoothing · Utah teapot camera pose optimization | fraction of poses recovered | Improved |
| use logistic distribution (ablation) | generalized stochastic smoothing · Warcraft shortest-path | exact match accuracy | Improved |
Closest papers by meaning
- Stochastic Variational Inference With Gradient Linearization (CVPR 2018 · similarity 0.87)
- Fast Stochastic Bregman Gradient Methods: Sharp Analysis and Variance Reduction (ICML 2021 · similarity 0.87)
- Marginalized Stochastic Natural Gradients for Black-Box Variational Inference (ICML 2021 · similarity 0.86)
- Stochastic Localization via Iterative Posterior Sampling (ICML 2024 · similarity 0.85)
- AdaptGrad: Adaptive Sampling to Reduce Noise (NeurIPS 2025 · similarity 0.85)
- Functional Gradient Flows for Constrained Sampling (NeurIPS 2024 · 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.
