Research map / Optimization / Implicit bias and generalization
Implicit regularization in Heavy-ball momentum 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 |
|---|---|---|---|
| Increase momentum from β=0.0, h=0.10 to β=0.8, h=0.02 at the same effective learning rate | SGD+M with DenseNet-121 · CIFAR-10 | test accuracy: 84.0±5.0 -> 88.6±0.7 | Helped |
| Increase momentum from β=0.0, h=0.10 to β=0.8, h=0.02 at the same effective learning rate | SGD+M with ResNet-18 · CIFAR-10 | test accuracy: 79.7±5.6 -> 82.4±0.4 | Helped |
| Increase momentum from β=0.0, h=0.10 to β=0.8, h=0.02 at the same effective learning rate | SGD+M with ResNet-50 · CIFAR-10 | test accuracy: 79.3±2.5 -> 82.4±0.7 | Helped |
| Increase momentum from β=0.0, h=0.10 to β=0.8, h=0.02 at the same effective learning rate | SGD+M with WideResNet-16-8 · CIFAR-10 | test accuracy: 65.3±18.1 -> 75.4±2.8 | Helped |
| Increase momentum from β=0.0, h=0.10 to β=0.8, h=0.02 at the same effective learning rate | SGD+M with DenseNet-121 · CIFAR-100 | test accuracy: 60.4±4.9 -> 64.7±0.8 | Helped |
| Increase momentum from β=0.0, h=0.10 to β=0.8, h=0.02 at the same effective learning rate | SGD+M with ResNet-18 · CIFAR-100 | test accuracy: 53.1±0.6 -> 54.3±0.6 | Helped |
| Increase momentum from β=0.0, h=0.10 to β=0.8, h=0.02 at the same effective learning rate | SGD+M with ResNet-50 · CIFAR-100 | test accuracy: 47.4±2.1 -> 49.6±0.7 | Helped |
| Increase momentum from β=0.0, h=0.10 to β=0.8, h=0.02 at the same effective learning rate | SGD+M with WideResNet-16-8 · CIFAR-100 | test accuracy: 38.6±3.8 -> 40.6±1.2 | Helped |
| Use β=0.8, h=0.02 instead of β=0.5, h=0.05 | SGD+M with ResNet-50 · CIFAR-100 | test accuracy: 50.3±1.0 -> 49.6±0.7 | Hurt |
| Add momentum and tune β at fixed h=0.1 | GD with ResNet-18 · MNIST, 1000 training images | test accuracy: 95.72% -> 96.22% | Helped |
Closest papers by meaning
- Does Momentum Change the Implicit Regularization on Separable Data? (NeurIPS 2022 · similarity 0.92)
- On the Origin of Implicit Regularization in Stochastic Gradient Descent (ICLR 2021 · similarity 0.91)
- Implicit Regularization or Implicit Conditioning? Exact Risk Trajectories of SGD in High Dimensions (NeurIPS 2022 · similarity 0.90)
- Accelerated Convergence of Stochastic Heavy Ball Method under Anisotropic Gradient Noise (ICLR 2024 · similarity 0.90)
- Effects of Momentum in Implicit Bias of Gradient Flow for Diagonal Linear Networks (AAAI 2025 · similarity 0.89)
- Perfect Parallelization in Mini-Batch SGD with Classical Momentum Acceleration (NeurIPS 2026 · similarity 0.89)
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.