hamrel-cxu/SVI-NN-training

Implementation for "An alternative approach to train neural networks using monotone variational inequality"

SVI NN training

Official implementation for An alternative approach to train neural networks using monotone variational inequality.

Please direct all implementation-related inquiries to cxu310@gatech.edu.

The released code compares SVI vs. SGD on the large OGB node classification task

• Training: SVI_OGB_128width, SVI_OGB_256width, SVI_OGB_512width
• Evaluation: SVI_OGB_results

Citation:

@article{xu2022alternative,
  title={An alternative approach to train neural networks using monotone variational inequality},
  author={Xu, Chen and Cheng, Xiuyuan and Xie, Yao},
  journal={arXiv preprint arXiv:2202.08876},
  year={2022}
}