LRBench

Introduction

A learning rate benchmarking and recommending tool, which will help practitioners efficiently select and compose good learning rate policies.

• Semi-automatic Learning Rate Tuning
• Evaluation: A set of Useful Metrics, covering Utility, Cost, and Robustness.
• Verification: Near-optimal Learning Rate

If you find this tool useful, please cite the following paper:

@ARTICLE{lrbench2019,
  author = {{Wu}, Yanzhao and {Liu}, Ling and {Bae}, Juhyun and {Chow}, Ka-Ho and
  {Iyengar}, Arun and {Pu}, Calton and {Wei}, Wenqi and {Yu}, Lei and
  {Zhang}, Qi},
  title = "{Demystifying Learning Rate Polices for High Accuracy Training of Deep Neural Networks}",
  journal = {arXiv e-prints},
  keywords = {Computer Science - Machine Learning, Statistics - Machine Learning},
  year = "2019",
  month = "Aug",
  eid = {arXiv:1908.06477},
  pages = {arXiv:1908.06477},
  archivePrefix = {arXiv},
  eprint = {1908.06477},
  primaryClass = {cs.LG},
  adsurl = {https://ui.adsabs.harvard.edu/abs/2019arXiv190806477W},
  adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}

Problem

Installation

Supported Platforms

Development / Contributing

Issues

Status

Contributors

See the people page for the full listing of contributors.

License

Copyright (c) 20XX-20XX Georgia Tech DiSL
Licensed under the Apache License.