Research map / Optimization / Nonconvex and smooth optimization
Towards Noise-adaptive, Problem-adaptive (Accelerated) Stochastic Gradient Descent
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 exponentially decreasing step-sizes instead of constant step-size for SGD | SGD · strongly-convex supervised learning (mushrooms, ijcnn, rcv1) | gradient norm stability | Helped |
| Replace stochastic line-search with decorrelated conservative SLS (line-search on previously sampled function) | SGD and ASGD with exponential step-size · mushrooms, ijcnn, rcv1 | gradient norm vs gradient evaluations | No clear effect |
| Standard SLS (correlated with sampled function) for step-size | SGD with exponential step-size · strong convexity theory, interpolation-free setting | convergence | Hurt |
| Use exponentially decreasing step-sizes with Nesterov acceleration (ASGD) versus non-accelerated SGD | SGD · smooth strongly-convex theory | convergence rate | Helped |
| Misestimate smoothness constant by factor ν>1 (offline estimate) | SGD with exponential step-size · smooth strongly-convex theory | convergence rate | Hurt |
Closest papers by meaning
- Two Sides of One Coin: the Limits of Untuned SGD and the Power of Adaptive Methods (NeurIPS 2023 · similarity 0.93)
- Convergence of Clipped SGD on Convex $(L_0,L_1)$-Smooth Functions (NeurIPS 2025 · similarity 0.92)
- Complexity Lower Bounds of Adaptive Gradient Algorithms for Non-convex Stochastic Optimization under Relaxed Smoothness (ICLR 2025 · similarity 0.92)
- Derivatives of Stochastic Gradient Descent in parametric optimization (NeurIPS 2024 · similarity 0.92)
- Controlling the Flow: Stability and Convergence for Stochastic Gradient Descent with Decaying Regularization (NeurIPS 2025 · similarity 0.91)
- Fast Last-Iterate Convergence of SGD in the Smooth Interpolation Regime (NeurIPS 2025 · similarity 0.91)
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.