Research map / Optimization / Implicit bias and generalization
Generalization Properties and Implicit Regularization for Multiple Passes SGM
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 |
|---|---|---|---|
| multiple passes with early stopping (SIGM, fixed step) vs one pass SGM with tuned step-size | SIGM constant step 1/sqrt(m) · Adult n=32562 / BreastCancer / Ijcnn1 n=49990 | test error (hinge loss) | Mixed |
| decaying step-size with multiple passes (SIGM, theta=1/2) instead of constant step | SIGM decaying step 1/4, theta=1/2 · Adult / BreastCancer / Ijcnn1 (all n reported) | test error (hinge loss) | Hurt |
| multiple passes with early stopping (SIGM, constant step) vs one pass SGM in training time | SIGM constant step · Adult n=32562 | training time (s) | Hurt |
| SGM and SIGM vs LIBSVM training time on large datasets | SGM/SIGM · Adult n=32562 | training time (s) | Helped |
| step-size decay exponent theta (fixed step-size theta=0 vs larger theta) as regularizer in one-pass SGM (ablation) | SGM with decaying step-size eta_t = 1/4 t^-theta · Adult n=1000 | test error (hinge loss) | Mixed |
| large fixed step-size (eta=1) in SGM vs small step (1e-3) (ablation) | SGM fixed step-size · Adult n=1000 | test error (hinge loss) | Mixed |
| early stopping (number of passes) as regularizer in SIGM with fixed step 1/sqrt(n) (ablation) | SIGM fixed step 1/sqrt(n) · Adult n=1000 | test error (hinge loss) | Helped |
Closest papers by meaning
- Optimal Learning for Multi-pass Stochastic Gradient Methods (NeurIPS 2016 · similarity 0.90)
- Implicit Regularization or Implicit Conditioning? Exact Risk Trajectories of SGD in High Dimensions (NeurIPS 2022 · similarity 0.90)
- Train faster, generalize better: Stability of stochastic gradient descent (ICML 2016 · similarity 0.89)
- Thinking Outside the Ball: Optimal Learning with Gradient Descent for Generalized Linear Stochastic Convex Optimization (NeurIPS 2022 · similarity 0.89)
- Controlling the Flow: Stability and Convergence for Stochastic Gradient Descent with Decaying Regularization (NeurIPS 2025 · similarity 0.88)
- Rapid Overfitting of Multi-Pass SGD in Stochastic Convex Optimization (ICML 2025 · 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.