Research map / Optimization / Optimizers
Adaptive Gradient Methods with Dynamic Bound of Learning Rate
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 dynamically bounded learning rates with AdaBound | Adam · CIFAR-10, DenseNet-121 | test accuracy: approximately 2% | Helped |
| Use dynamically bounded learning rates with AdaBound | Adam · CIFAR-10, ResNet-34 | test accuracy: 1% | Helped |
| Use dynamically bounded learning rates with AdaBound | Adam · Penn Treebank, 1-layer LSTM | perplexity: 1.1% | Helped |
| Use dynamically bounded learning rates with AdaBound | Adam · Penn Treebank, 3-layer LSTM | perplexity: over 2.8% | Helped |
| Use dynamically bounded learning rates with AdaBound | Adam · MNIST, 1-layer perceptron | test accuracy | Helped |
| Use dynamically bounded learning rates with AdaBound or AMSBound | Adam or AMSGrad · CIFAR-10, DenseNet-121 and ResNet-34 | test accuracy: 2%; 1% | Helped |
| Vary the convergence speed of the dynamic bound functions (ablation) | AdaBound · CIFAR-10, ResNet-34 | test accuracy | No clear effect |
| Vary the final step-size target α* (ablation) | AdaBound · CIFAR-10, ResNet-34 | test accuracy | No clear effect |
| Vary the step size α (ablation) | SgdM · CIFAR-10, ResNet-34 | test accuracy: lower than 70% | Mixed |
| Use AdaBound instead of SgdM at matching step-size settings | SgdM · CIFAR-10, ResNet-34 | test accuracy | Helped |
| Use a gradual dynamic-bound transition rather than an abrupt switch from Adam to SGD | Adam-to-SGD optimization · General | test accuracy | Helped |
| Use dynamic bounds with AMSBound | AMSGrad · CIFAR-10 and Penn Treebank | test accuracy or perplexity | Helped |
Closest papers by meaning
- The Marginal Value of Adaptive Gradient Methods in Machine Learning (NeurIPS 2017 · similarity 0.91)
- A Comprehensive Framework for Analyzing the Convergence of Adam: Bridging the Gap with SGD (ICML 2025 · similarity 0.89)
- Prodigy: An Expeditiously Adaptive Parameter-Free Learner (ICML 2024 · similarity 0.89)
- Adaptive Methods through the Lens of SDEs: Theoretical Insights on the Role of Noise (ICLR 2025 · similarity 0.88)
- Surge Phenomenon in Optimal Learning Rate and Batch Size Scaling (NeurIPS 2024 · similarity 0.87)
- Remove that Square Root: A New Efficient Scale-Invariant Version of AdaGrad (NeurIPS 2024 · 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.