Research map / Optimization / Optimizers
AGD: an Auto-switchable Optimizer using Stepwise Gradient Difference for Preconditioning Matrix
Yun Yue, Zhiling Ye, Jiadi Jiang, Yongchao Liu, Ke Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| AGD instead of AdaBelief | AdaBelief · PTB 1-layer LSTM | test perplexity: 84.40 -> 81.23 | Improved |
| AGD instead of AdaBelief | AdaBelief · PTB 3-layer LSTM | test perplexity: 61.34 -> 60.89 | Improved |
| AGD instead of AdaBelief | AdaBelief · IWSLT14 De-En | BLEU: 35.93 -> 35.94 | No clear effect |
| AGD instead of Adam | Adam · ImageNet ResNet18 | top-1 accuracy: 64.03 -> 70.35 | Improved |
| AGD instead of SGD | SGD · ImageNet ResNet18 | top-1 accuracy: 69.94 -> 70.35 | Improved |
| AGD instead of SGD | SGD · CIFAR-10 ResNet32 | top-1 accuracy: 93.10 -> 93.12 | No clear effect |
| AGD instead of Adam | Adam · Avazu MLP | test AUC: 0.7458 -> 0.7480 | Improved |
| AGD instead of Adam | Adam · Criteo DCN | test AUC: 0.8023 -> 0.8023 | No clear effect |
| AGD instead of SGD | SGD · Criteo DCN | test AUC: 0.7296 -> 0.8023 | Improved |
| add AMSGrad condition to AGD (ablation) | AGD · CIFAR-10 ResNet20 | top-1 accuracy: 92.35 -> 92.25 | Hurt |
| AGD with stepwise-gradient-difference preconditioner instead of AdaHessian (cost) | AdaHessian · IWSLT14 Transformer small | time per epoch: 750 -> 278 | Improved |
| AGD instead of AdamW (cost) | AdamW · IWSLT14 Transformer small | time per epoch: 260 -> 278 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AdaHessian — ADAHESSIAN: An Adaptive Second Order Optimizer for Machine Learning (AAAI 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SGD with Adaptive Preconditioning: Unified Analysis and Momentum Acceleration (ICLR 2026 · similarity 0.88)
- On The Concurrence of Layer-wise Preconditioning Methods and Provable Feature Learning (ICML 2025 · similarity 0.87)
- Adam Reduces a Unique Form of Sharpness: Theoretical Insights Near the Minimizer Manifold (NeurIPS 2025 · similarity 0.86)
- AdaFisher: Adaptive Second Order Optimization via Fisher Information (ICLR 2025 · similarity 0.86)
- Exact, Tractable Gauss-Newton Optimization in Deep Reversible Architectures Reveal Poor Generalization (NeurIPS 2024 · similarity 0.86)
- Can We Remove the Square-Root in Adaptive Gradient Methods? A Second-Order Perspective (ICML 2024 · similarity 0.85)
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.
