Research map / Optimization / Optimizers
Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks
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 the baseline optimizer with Neumann | Inception-V3 · ImageNet | Top-1 validation error: 21.7 % -> 20.8 % | Helped |
| Replace the baseline optimizer with Neumann | ResNet-50 · ImageNet | Top-1 validation error: 23.9 % -> 23.0 % | Helped |
| Replace the baseline optimizer with Neumann | ResNet-101 · ImageNet | Top-1 validation error: 22.6 % -> 21.7% | Helped |
| Replace the baseline optimizer with Neumann | Inception-ResNet-V2 · ImageNet | Top-1 validation error: 20.3 % -> 19.5 % | Helped |
| Add cubic and repulsive regularizers (ablation) | Neumann on ResNet-50 · ImageNet, batch size 4000 | Top-1 error: 23.5 % -> 23.0 % | Helped |
| Use Neumann without cubic and repulsive regularizers instead of the baseline optimizer (ablation) | ResNet-50 · ImageNet, batch size 4000 | Top-1 error: 24.3 % -> 23.5 % | Helped |
| Use Neumann on Tacotron | Tacotron · Sequence-to-sequence speech synthesis | Speedup or model quality | No clear effect |
| Use fewer Neumann training steps while targeting baseline validation error | ImageNet models · ImageNet | Training steps: 10-30\% decrease | Helped |
Closest papers by meaning
- Descending through a Crowded Valley - Benchmarking Deep Learning Optimizers (ICML 2021 · similarity 0.91)
- DeepOBS: A Deep Learning Optimizer Benchmark Suite (ICLR 2019 · similarity 0.91)
- Learned Optimizers that Scale and Generalize (ICML 2017 · similarity 0.90)
- MARS: Unleashing the Power of Variance Reduction for Training Large Models (ICML 2025 · similarity 0.88)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.88)
- Adam-mini: Use Fewer Learning Rates To Gain More (ICLR 2025 · similarity 0.88)
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.