Research map / Optimization / Optimizers
Towards Dynamic Interleaving Optimizers
Yile Chen, Zeyi Wen, Jian Chen, Jin Huang
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 |
|---|---|---|---|
| switch optimizers dynamically with DOIT instead of a static optimizer | ViT full training · cifar10 | accuracy (%): 97.58 -> 98.04 | Improved |
| use DOIT on a complex imbalanced dataset | full training · imagenet-a | top-1 accuracy (%): 16.31 -> 18.47 | Improved |
| use DOIT instead of hybrid optimizers (SWATS, Padam, AdaBound) | RoBERTa LoRA · qqp | accuracy (%): 73.04 -> 85.57 (vs Padam) | Improved |
| use DOIT on additional tasks | translation, regression, detection · wmt14 / eunite / coco | BLEU / R2 / AP50: 28.24 -> 28.68 BLEU; 71.81 -> 73.12 R2; 57.44 -> 59.21 AP50 | Improved |
| use DOIT instead of LR schedulers (stepLR, cosLR) | full training · mnist | accuracy (%): 99.46 -> 99.51 | No clear effect |
| add DOIT scheduling overhead | training time · ResNet18 cifar10 | time overhead | No clear effect |
| replace DOIT components (random/periodic switching, no transferability weight, uniform init, random projection/UMAP) (ablation) | DOIT · ablation suite | accuracy | Improved |
| expand optimizer/hyperparameter search space (ablation) | DOIT · usps | accuracy and time | No clear effect |
Closest papers by meaning
- Descending through a Crowded Valley - Benchmarking Deep Learning Optimizers (ICML 2021 · similarity 0.87)
- DDPNOpt: Differential Dynamic Programming Neural Optimizer (ICLR 2021 · similarity 0.87)
- Understanding and correcting pathologies in the training of learned optimizers (ICML 2019 · similarity 0.86)
- Learned Optimizers that Scale and Generalize (ICML 2017 · similarity 0.86)
- Neural Network Optimization Reimagined: Decoupled Techniques for Scratch and Fine-Tuning (TPAMI 2026 · similarity 0.86)
- Investigation into Training Dynamics of Learned Optimizers (Student Abstract) (AAAI 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.
