GraSS: Scalable Influence Function with Sparse Gradient Compression

Posted: May 25th 2025
Abstract

We propose an efficient gradient compression algorithm to accelerate and scale influence function to billion-scale models.

GraSS: Scalable Influence Function with Sparse Gradient Compression

arXiv | GitHub

Brief Summary

TBA

Citation

@misc{hu2025grass,
  author        = {Pingbang Hu and Joseph Melkonian and Weijing Tang and Han Zhao and Jiaqi W. Ma},
  title         = {GraSS: Scalable Influence Function with Sparse Gradient Compression},
  archiveprefix = {arXiv},
  eprint        = {2505.18976},
  primaryclass  = {cs.LG},
  url           = {https://arxiv.org/abs/2505.18976},
  year          = {2025}
}