Research map / Architectures and efficiency / RNNs and normalization
DeepOBS: A Deep Learning Optimizer Benchmark Suite
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 Momentum instead of SGD or Adam | deep learning optimization · CIFAR-100 All-CNN-C | test accuracy: 57.06 % -> 60.33 %; 56.15 % -> 60.33 % | Helped |
| Use Adam instead of SGD or Momentum | deep learning optimization · CIFAR-10 CNN | test accuracy: 83.71 % -> 84.75 %; 84.41 % -> 84.75 % | Helped |
| Use SGD instead of Momentum or Adam | deep learning optimization · Tolstoi CharRNN | test accuracy: 61.30 % -> 62.07 %; 61.23 % -> 62.07 % | Helped |
| Choose optimizer by benchmark problem rather than applying one optimizer universally | SGD, Momentum, and Adam · eight DeepOBS benchmark problems | test performance | Mixed |
| Tune learning rate separately for each problem | SGD, Momentum, and Adam · eight DeepOBS benchmark problems | test performance | Helped |
| Use Adam rather than SGD or Momentum | deep learning optimization · most DeepOBS test problems | test performance | Mixed |
| Compare optimizers using multiple random-seed runs | optimizer evaluation · stochastic optimization benchmarks | performance variability | Helped |
| Report training and test loss and accuracy, rather than a single metric | optimizer evaluation · deep learning benchmarks | generalization assessment | Helped |
Closest papers by meaning
- Descending through a Crowded Valley - Benchmarking Deep Learning Optimizers (ICML 2021 · similarity 0.92)
- Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks (ICLR 2018 · similarity 0.91)
- Learning Gradient Descent: Better Generalization and Longer Horizons (ICML 2017 · similarity 0.90)
- Optimizer Benchmarking Needs to Account for Hyperparameter Tuning (ICML 2020 · similarity 0.89)
- Diving into the shallows: a computational perspective on large-scale shallow learning (NeurIPS 2017 · similarity 0.89)
- Benchmarking Predictive Coding Networks -- Made Simple (ICLR 2025 · 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.