Research map / Optimization / Optimizers
ADAHESSIAN: An Adaptive Second Order Optimizer for Machine Learning
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 |
|---|---|---|---|
| Replace Adam's gradient-based second moment with AdaHessian's estimated Hessian diagonal | ResNet20 · CIFAR-10 | accuracy: 1.80% | Helped |
| Replace Adam's gradient-based second moment with AdaHessian's estimated Hessian diagonal | ResNet32 · CIFAR-10 | accuracy: 1.45% | Helped |
| Use AdaHessian instead of Adam | ResNet18 · ImageNet | accuracy: 5.53% | Helped |
| Use AdaHessian instead of AdamW | Transformer small · IWSLT14 De-En | BLEU: 0.13 | Helped |
| Use AdaHessian instead of AdamW | Transformer base · WMT14 En-De | BLEU: 0.33 | Helped |
| Use AdaHessian instead of AdamW | Tensorized Transformer · PTB | PPL: 2.7 | Helped |
| Use AdaHessian instead of AdamW | Tensorized Transformer · Wikitext-103 | PPL: 1.0 | Helped |
| Use AdaHessian instead of AdamW* | SqueezeBERT · GLUE development average | average score: 0.41 | Helped |
| Add Hessian momentum to the Hessian diagonal estimate (ablation) | AdaHessian · Noisy 1D quadratic example | iterations to converge: more than 1000 -> 7 | Helped |
| Vary spatial averaging block size from 1 to 128 (ablation) | AdaHessian · IWSLT14 De-En | BLEU: 35.67 $\pm$ .10 to 35.67 $\pm$ .11 | No clear effect |
| Scale learning rate by 10 (ablation) | AdaHessian · IWSLT14 De-En | BLEU: 35.79 $\pm$ .06 -> 32.48 $\pm$ .83 | Hurt |
| Reduce Hutchinson Hessian computation frequency from every iteration to every 2 iterations (ablation) | AdaHessian on ResNet20 · CIFAR-10 | accuracy: 92.13 $\pm$ .08 -> 92.40 $\pm$ .04 | Helped |
Closest papers by meaning
- FOSI: Hybrid First and Second Order Optimization (ICLR 2024 · similarity 0.88)
- Advancing the Lower Bounds: an Accelerated, Stochastic, Second-order Method with Optimal Adaptation to Inexactness (ICLR 2024 · similarity 0.86)
- Global curvature for second-order optimization of neural networks (ICML 2025 · similarity 0.86)
- Efficient Curvature-Aware Hypergradient Approximation for Bilevel Optimization (ICML 2025 · similarity 0.85)
- AdaFisher: Adaptive Second Order Optimization via Fisher Information (ICLR 2025 · similarity 0.85)
- Remove that Square Root: A New Efficient Scale-Invariant Version of AdaGrad (NeurIPS 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.