Research map / Probabilistic and scientific modeling / Bayesian optimization
AutoLRS: Automatic Learning-Rate Schedule by Bayesian Optimization on the Fly
Yuchen Jin, Tianyi Zhou, Liangyu Zhao, Yibo Zhu, Chuanxiong Guo, Marco Canini, Arvind Krishnamurthy
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 |
|---|---|---|---|
| use stage-wise Bayesian optimization with exponential loss forecasting instead of the original hand-tuned learning-rate schedule | ResNet-50 · ImageNet | training-step speedup: 1.22× | Improved |
| use stage-wise Bayesian optimization with exponential loss forecasting instead of the original hand-tuned learning-rate schedule | Transformer base · WMT 2014 English-German | training-step speedup: 1.43× | Improved |
| use stage-wise Bayesian optimization with exponential loss forecasting instead of the original hand-tuned learning-rate schedule | BERT_BASE · BERT pre-training | training-step speedup: 1.5× | Improved |
| add exponential forecasting to predict loss at the full stage horizon rather than using loss at τ′ (ablation) | AutoLRS · VGG-16, CIFAR-10 | top-1 accuracy: 91.73%, 92.59%, 92.24% | Improved |
| use Bayesian optimization instead of random search (ablation) | AutoLRS with exponential forecasting · VGG-16, CIFAR-10 | average top-1 accuracy: 0.09% | Improved |
| use Bayesian optimization instead of grid search (ablation) | AutoLRS with exponential forecasting · VGG-16, CIFAR-10 | average top-1 accuracy: 0.24% | Improved |
| use Bayesian optimization instead of grid search (ablation) | AutoLRS with exponential forecasting · VGG-16, CIFAR-100 | mean top-1 accuracy: 73.05%±0.14%; 72.63%±0.56% | Improved |
| use AutoLRS instead of the hand-tuned schedule | BERT_BASE · MRPC Dev | accuracy: 86.5 -> 88.0 | Improved |
| use AutoLRS instead of the hand-tuned schedule | BERT_BASE · MNLI Dev | accuracy: 82.2 -> 82.5 | Improved |
| use AutoLRS instead of the hand-tuned schedule | BERT_BASE · CoLA Dev | accuracy: 47.8 -> 47.6 | Hurt |
| use AutoLRS instead of the hand-tuned schedule | BERT_BASE · SQuAD v1.1 Dev | F1: 87.0 -> 87.1 | Improved |
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.
- Hyperband — Hyperband: A Novel Bandit-Based Approach to Hyperparameter Optimization (JMLR 2018) · better on 2 of 2 shared results
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
- AdaLRS: Loss-Guided Adaptive Learning Rate Search for Efficient Foundation Model Pretraining (NeurIPS 2025 · similarity 0.87)
- Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights (NeurIPS 2025 · similarity 0.86)
- Scaling Law with Learning Rate Annealing (NeurIPS 2025 · similarity 0.86)
- A Multi-Power Law for Loss Curve Prediction Across Learning Rate Schedules (ICLR 2025 · similarity 0.86)
- Benchmarking Optimizers for Large Language Model Pretraining (NeurIPS 2026 · similarity 0.85)
- The Surprising Agreement Between Convex Optimization Theory and Learning-Rate Scheduling for Large Model Training (ICML 2025 · 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.
