What type of study is this?

This is a Experimental Study study.

September 29, 2025Open Access

GraSS: Scalable Influence Function with Sparse Gradient Compression

View Full Paper

Ask AI

Bookmark

GraSS achieves substantial speedups in per-sample gradient computation while maintaining data influence fidelity.
FactGraSS specifically enhances throughput by achieving up to 165% faster performance on billion-scale models.
The method leverages sparsity in per-sample gradients to improve computational efficiency significantly.
Extensive experiments validate the effectiveness of GraSS compared to prior state-of-the-art baselines.

No takes yet. Share an insight, caveat, or question.

Proposed GraSS algorithm reduces space complexity in gradient computation, highlighting scalability and efficiency.

Hu et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark