Research map / Optimization / Optimizers
Beating SGD Saturation with Tail-Averaging and Minibatching
Nicole Muecke, Gergely Neu, Lorenzo Rosasco
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 non-trivial tail averaging instead of uniform averaging (ablation) | SGD · Least squares learning with r>1/2 | excess risk rate: \left(\frac{1}{\gamma T}\right)^{2 \min(r,1/2)+1} -> \left(\frac{1}{\gamma T}\right)^{2r+1} | Improved |
| Increase smoothness r beyond 1/2 while using uniform averaging (ablation) | Uniformly averaged SGD · Synthetic data, excess risk versus r | excess risk | No clear effect |
| Increase smoothness r beyond 1/2 while using tail averaging (ablation) | Tail-averaged SGD · Synthetic data, excess risk versus r | excess risk | Improved |
| Increase minibatch size while tail averaging (ablation) | Tail-averaged SGD · Synthetic data, single pass | excess risk | Improved |
| Use tail averaging with a constant stepsize and smaller minibatch than the compared multi-pass method (ablation) | Minibatch SGD · Least squares learning | excess risk rate: n^{-\frac{2r+1}{2r+1+\nu}} | Improved |
Closest papers by meaning
- Asymptotic theory of SGD with a general learning-rate (NeurIPS 2025 · similarity 0.90)
- Parallelizing Stochastic Gradient Descent for Least Squares Regression: Mini-batching, Averaging, and Model Misspecification (JMLR 2018 · similarity 0.90)
- On the Convergence of mSGD and AdaGrad for Stochastic Optimization (ICLR 2022 · similarity 0.89)
- Direction Matters: On the Implicit Bias of Stochastic Gradient Descent with Moderate Learning Rate (ICLR 2021 · similarity 0.88)
- Towards Stability and Generalization Bounds in Decentralized Minibatch Stochastic Gradient Descent (AAAI 2024 · similarity 0.88)
- Tighter Convergence Bounds for Shuffled SGD via Primal-Dual Perspective (NeurIPS 2024 · similarity 0.88)
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.
