Research map / Optimization / Optimizers
A Statistical Online Inference Approach in Averaged Stochastic Approximation
Chuhan Xie, Zhihua Zhang
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 random scaling instead of batch means | constant step size SGD inference · Linear regression d=5, eta=0.01, n=5,000 | coverage rate (%): 69.8(1.451) -> 92.8(0.817) | Improved |
| use random scaling instead of batch means | constant step size SGD inference · Linear regression d=5, eta=0.01, n=40,000 | coverage rate (%): 86.7(1.073) -> 94.9(0.695) | Improved |
| use random scaling instead of spectral variance | constant step size SGD inference · Linear regression d=5, eta=0.01, n=5,000 | coverage rate (%): 88.6(1.005) -> 92.8(0.817) | Improved |
| use random scaling instead of spectral variance | constant step size SGD inference · Linear regression d=5, eta=0.1, n=40,000 | coverage rate (%): 92.7(0.822) -> 93.0(0.806) | No clear effect |
| use random scaling instead of batch means | constant step size SGD inference · Linear regression d=5, eta=0.1, n=40,000 | coverage rate (%): 92.8(0.817) -> 93.0(0.806) | No clear effect |
| use random scaling instead of batch means | constant step size SGD inference · Linear regression d=5, eta=0.01, n=40,000 | CI length: 0.015(0.001) -> 0.026(0.010) | Hurt |
| use random scaling instead of spectral variance | constant step size SGD inference · Linear regression d=5, eta=0.01, n=40,000 | CI length: 0.018(0.003) -> 0.026(0.010) | Hurt |
| use random scaling instead of batch means | constant step size SGD inference · Logistic regression d=5, eta=0.01, n=40,000 | coverage rate for beta*_1 (%): 55.9(1.570) -> 95.1(0.682) | Improved |
| use random scaling instead of spectral variance | constant step size SGD inference · Logistic regression d=5, eta=0.01, n=40,000 | coverage rate for beta*_1 (%): 89.7(0.961) -> 95.1(0.682) | Improved |
| increase step size from 0.01 to 0.1 for batch means | batch means inference · Linear regression d=5, n=5,000 | coverage rate (%): 69.8(1.451) -> 92.9(0.812) | Improved |
| increase step size from 0.01 to 0.1 for random scaling | random scaling inference · Linear regression d=5, n=5,000 | coverage rate (%): 92.8(0.817) -> 94.3(0.733) | No clear effect |
| increase sample size from n=5,000 to n=40,000 for batch means | batch means inference · Linear regression d=5, eta=0.01 | coverage rate (%): 69.8(1.451) -> 86.7(1.073) | Improved |
Closest papers by meaning
- Effectiveness of Constant Stepsize in Markovian LSA and Statistical Inference (AAAI 2024 · similarity 0.89)
- A General-Purpose Theorem for High-Probability Bounds of Stochastic Approximation with Polyak Averaging (NeurIPS 2025 · similarity 0.87)
- Statistical inference for Linear Stochastic Approximation with Markovian Noise (NeurIPS 2025 · similarity 0.87)
- Bayesian Online Natural Gradient (BONG) (NeurIPS 2024 · similarity 0.86)
- The Collusion of Memory and Nonlinearity in Stochastic Approximation With Constant Stepsize (NeurIPS 2024 · similarity 0.86)
- Statistical Guarantees for High-Dimensional Stochastic Gradient Descent (NeurIPS 2025 · similarity 0.86)
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.
