Research map / Optimization / Optimizers
Fast and Robust Online Inference with Stochastic Gradient Descent via Random Scaling
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 random scaling instead of batch-means inference | Online SGD confidence intervals for logistic regression · Logistic regression d=200, n=100000, γ0=0.5, a=0.505 | coverage: 0.644 -> 0.919 | Helped |
| Use random scaling instead of plug-in inference | Online SGD confidence intervals for logistic regression · Logistic regression d=200, n=100000, γ0=0.5, a=0.505 | computation time (sec.): 955.0 -> 170.3 | Helped |
| Use random scaling instead of batch-means inference | Online SGD confidence intervals for linear regression · Linear regression d=5, n=100000, γ0=0.5, a=0.505 | coverage: 0.893 -> 0.957 | Helped |
| Increase the learning-rate exponent to a=0.667 | Random Scale online inference · Linear regression d=5, γ0=0.5, n=100000 | coverage: 0.957 -> 0.946 | No clear effect |
| Increase the learning-rate exponent to a=0.667 | Batch-mean online inference · Linear regression d=5, γ0=0.5, n=100000 | coverage: 0.893 -> 0.802 | Hurt |
| Increase the dimension to d=200 | Random Scale online inference · Logistic regression, n=100000, γ0=0.5, a=0.505 | computation time (sec.): 8.4 -> 170.3 | Hurt |
| Update only the target parameter's scaling element rather than the full matrix | Random Scale online inference · Logistic regression d=800, n=10^5, γ0=0.5, a=0.505 | computation time (sec.): 11.7 seconds at d=800 | Helped |
Closest papers by meaning
- Statistical Guarantees for High-Dimensional Stochastic Gradient Descent (NeurIPS 2025 · similarity 0.87)
- Bayesian Online Natural Gradient (BONG) (NeurIPS 2024 · similarity 0.85)
- Statistical Inference for Gradient Boosting Regression (NeurIPS 2025 · similarity 0.85)
- Gradient Descent Converges Arbitrarily Fast for Logistic Regression via Large and Adaptive Stepsizes (ICML 2025 · similarity 0.85)
- Near-Optimal Streaming Heavy-Tailed Statistical Estimation with Clipped SGD (NeurIPS 2024 · similarity 0.85)
- Nonasymptotic Analysis of Stochastic Gradient Descent with the Richardson–Romberg Extrapolation (ICLR 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.