Research map / Optimization / Nonconvex and smooth optimization
On Acceleration with Noise-Corrupted Gradients
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 AGD+ instead of AXGD | Accelerated optimization with noisy gradient oracles · Gradient oracle queries | oracle complexity: by a factor of two | Helped |
| Apply restart and slow-down when the weighted gradient signal is dominated by estimated noise (ablation) | Accelerated methods including AGD+ · Unconstrained smooth minimization, high gradient noise | mean approximation error | Helped |
| Apply restart and slow-down when the weighted gradient signal is dominated by estimated noise (ablation) | Accelerated methods including AGD+ · Unconstrained smooth minimization, high gradient noise | variance of approximation error | Helped |
| Apply restart and slow-down (ablation) | Accelerated methods · Noiseless smooth minimization | approximation error | No clear effect |
| Apply restart and slow-down (ablation) | Accelerated methods · Unconstrained smooth minimization, low gradient noise | approximation error | No clear effect |
| Constrain optimization to a bounded feasible region | Accelerated methods with noisy gradients · Hard instance over the probability simplex | noise-induced error accumulation | Helped |
| Apply restart and slow-down (ablation) | Accelerated methods · Bounded-diameter constrained minimization | mean noise error | No clear effect |
| Apply restart and slow-down (ablation) | Accelerated methods · Bounded-diameter constrained minimization, high noise variance | error variance | Helped |
| Use AGD instead of AXGD or AGD+ | Accelerated optimization · LASSO on Epileptic Seizure Recognition Dataset, noiseless and low noise | convergence speed | Helped |
| Use AGD instead of AXGD or AGD+ | Accelerated optimization · LASSO on Epileptic Seizure Recognition Dataset, high noise | error variance | Hurt |
| Use AGD instead of AXGD or AGD+ | Accelerated optimization · LASSO on Epileptic Seizure Recognition Dataset, very high noise | mean approximation error | Hurt |
| Use restart and slow-down (ablation) | Accelerated methods · Logistic regression on Epileptic Seizure Recognition Dataset, high noise | mean and variance of approximation error | Helped |
Closest papers by meaning
- A Universally Optimal Multistage Accelerated Stochastic Gradient Method (NeurIPS 2019 · similarity 0.90)
- Nesterov acceleration despite very noisy gradients (NeurIPS 2024 · similarity 0.90)
- Accelerated Zeroth-order Method for Non-Smooth Stochastic Convex Optimization Problem with Infinite Variance (NeurIPS 2023 · similarity 0.89)
- Stochastic Optimization with Heavy-Tailed Noise via Accelerated Gradient Clipping (NeurIPS 2020 · similarity 0.88)
- Gradient correlation is a key ingredient to accelerate SGD with momentum (ICLR 2025 · similarity 0.87)
- Nesterov acceleration in benignly non-convex landscapes (ICLR 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.