Research map / Optimization / Optimizers
Statistical Inference Using SGD
Tianyang Li, Liu Liu, Anastasios Kyrillidis, Constantine Caramanis
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 η=0.1 rather than η=0.004 with t=2500 (ablation) | SGD inference · Linear regression synthetic, Experiment 1 | coverage probability: 0.916 -> 0.960 | Improved |
| Use η=0.1 rather than η=0.004 with t=2500 (ablation) | SGD inference · Linear regression synthetic, Experiment 1 | average confidence interval width: 3.70 -> 4.53 | Improved |
| Use t=2500 rather than t=100 at η=0.1 (ablation) | SGD inference · Linear regression synthetic, Experiment 1 | coverage probability: 0.957 -> 0.960 | No clear effect |
| Use η=0.1 rather than η=0.004 with t=2500 (ablation) | SGD inference · Linear regression synthetic, Experiment 2 | coverage probability: 0.897 -> 0.963 | Improved |
| Use η=0.1 rather than η=0.004 with t=2500 (ablation) | SGD inference · Logistic regression synthetic, Experiment 1 | coverage probability: 0.86 -> 0.939 | Improved |
| Use η=0.1 rather than η=0.004 with t=2500 (ablation) | SGD inference · Logistic regression synthetic, Experiment 2 | coverage probability: 0.851 -> 0.947 | Improved |
| Use η=0.1 rather than η=0.004 with t=2500 (ablation) | SGD inference · Logistic regression synthetic, Experiment 1 | average confidence interval width: 0.194 -> 0.258 | Mixed |
| Use η=0.1 rather than η=0.004 with t=2500 (ablation) | SGD inference · Logistic regression synthetic, Experiment 2 | average confidence interval width: 0.195 -> 0.266 | Mixed |
| Use SGD inference instead of bootstrap | SGD inference · Linear regression synthetic, Experiment 1 | coverage probability: 0.941 -> 0.960 | Improved |
| Use SGD inference instead of bootstrap | SGD inference · Linear regression synthetic, Experiment 2 | coverage probability: 0.938 -> 0.963 | Improved |
| Use SGD inference instead of bootstrap | SGD inference · Logistic regression synthetic, Experiment 1 | coverage probability: 0.932 -> 0.939 | Improved |
| Use SGD inference instead of normal approximation | SGD inference · Logistic regression synthetic, Experiment 1 | coverage probability: 0.957 -> 0.939 | Hurt |
Closest papers by meaning
- On the Utility of Equal Batch Sizes for Inference in Stochastic Gradient Descent (JMLR 2025 · similarity 0.91)
- A Variational Analysis of Stochastic Gradient Algorithms (ICML 2016 · similarity 0.91)
- Online stochastic gradient descent on non-convex losses from high-dimensional inference (JMLR 2021 · similarity 0.90)
- Stochastic Gradient Descent as Approximate Bayesian Inference (JMLR 2017 · similarity 0.89)
- Statistical Guarantees for High-Dimensional Stochastic Gradient Descent (NeurIPS 2025 · similarity 0.87)
- Demystifying SGD with Doubly Stochastic Gradients (ICML 2024 · similarity 0.87)
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.
