MSDF-SGD: Most-Significant Digit-First Stochastic Gradient Descent for Arbitrary-Precision Training

Changjun Song, Yongming Tang, Jiyuan Liu 0006, Sige Bian, Danni Deng, He Li 0008. MSDF-SGD: Most-Significant Digit-First Stochastic Gradient Descent for Arbitrary-Precision Training. In Nele Mentens, Leonel Sousa, Pedro Trancoso, Nikela Papadopoulou, Ioannis Sourdis, editors, 33rd International Conference on Field-Programmable Logic and Applications, FPL 2023, Gothenburg, Sweden, September 4-8, 2023. pages 159-165, IEEE, 2023. [doi]

@inproceedings{SongTLBDL23,
  title = {MSDF-SGD: Most-Significant Digit-First Stochastic Gradient Descent for Arbitrary-Precision Training},
  author = {Changjun Song and Yongming Tang and Jiyuan Liu 0006 and Sige Bian and Danni Deng and He Li 0008},
  year = {2023},
  doi = {10.1109/FPL60245.2023.00030},
  url = {https://doi.org/10.1109/FPL60245.2023.00030},
  researchr = {https://researchr.org/publication/SongTLBDL23},
  cites = {0},
  citedby = {0},
  pages = {159-165},
  booktitle = {33rd International Conference on Field-Programmable Logic and Applications, FPL 2023, Gothenburg, Sweden, September 4-8, 2023},
  editor = {Nele Mentens and Leonel Sousa and Pedro Trancoso and Nikela Papadopoulou and Ioannis Sourdis},
  publisher = {IEEE},
  isbn = {979-8-3503-4151-5},
}