Research map / Optimization / Optimizers
How Do Adam and Training Strategies Help BNNs Optimization
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 Adam instead of SGD | StrongBaseline BNN · ImageNet | Top-1 accuracy: 58.98 -> 61.49 | Helped |
| Use Adam instead of SGD | StrongBaseline BNN with two-step training · ImageNet | Top-1 accuracy: 61.25 -> 63.23 | Helped |
| Apply the proposed two-step training strategy | ReActNet-A · ImageNet | Top-1 accuracy: 69.4 -> 70.5 | Helped |
| Apply the proposed two-step training strategy | StrongBaseline · ImageNet | Top-1 accuracy: 60.9 -> 63.2 | Helped |
| Binarize activations before weights rather than weights before activations (ablation) | StrongBaseline two-step training · ImageNet | Top-1 accuracy: 60.17 -> 63.23 | Helped |
| Use weight decay 5e-6 rather than 0 in one-step training (ablation) | Adam, StrongBaseline · ImageNet | Top-1 accuracy: 61.49 -> 61.89 | Helped |
| Use weight decay -1e-4 rather than 0 in one-step training (ablation) | Adam, StrongBaseline · ImageNet | Top-1 accuracy: 61.49 -> 26.21 | Hurt |
| Use two-step weight decay, with 5e-6 in Step1 and 0 in Step2, rather than one-step weight decay 5e-6 (ablation) | Adam, StrongBaseline · ImageNet | Top-1 accuracy: 61.89 -> 63.23 | Helped |
| Use Adam instead of RMSprop (ablation) | StrongBaseline BNN · ImageNet | Top-1 accuracy: 57.90 -> 61.49 | Helped |
| Use Adam instead of AdaGrad (ablation) | StrongBaseline BNN · ImageNet | Top-1 accuracy: 50.74 -> 61.49 | Helped |
| Use Adam instead of AdaDelta (ablation) | StrongBaseline BNN · ImageNet | Top-1 accuracy: 56.90 -> 61.49 | Helped |
| Use Adam instead of AdaBound (ablation) | StrongBaseline BNN · ImageNet | Top-1 accuracy: 58.13 -> 61.49 | Helped |
Closest papers by meaning
- Training Binary Neural Networks via Gaussian Variational Inference and Low-Rank Semidefinite Programming (NeurIPS 2024 · similarity 0.88)
- Understanding the Generalization of Stochastic Gradient Adam in Learning Neural Networks (NeurIPS 2025 · similarity 0.87)
- A Comprehensive Framework for Analyzing the Convergence of Adam: Bridging the Gap with SGD (ICML 2025 · similarity 0.87)
- The Hidden Ratio in Adam: Stable Structure, Compression, and Sign Dynamics (NeurIPS 2026 · similarity 0.87)
- The Rich and the Simple: On the Implicit Bias of Adam and SGD (NeurIPS 2025 · similarity 0.86)
- Fast and Slow Gradient Approximation for Binary Neural Network Optimization (AAAI 2025 · similarity 0.86)
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.