The authors propose EDiS, a three-stage framework: edge scoring, structural extraction, and budgeted composition.
EDiS caches edge-disjoint subgraphs to cheaply vary sparse GNN training graphs
Kaplan et al. decompose a graph once into cacheable, non-overlapping subgraphs and recombine them per epoch, achieving top benchmark scores across 19 datasets.
Big Tech
Sai Karthik Navuluru · Siddhartha Shankar Das · Franck Dernoncourt · S M Ferdous · Ryan A. Rossi · Nesreen K. Ahmed · +4 more
University of Texas at Dallas · Pacific Northwest National Laboratory · Adobe Research · University of North Carolina at Charlotte · Cisco
Research Digest··3 min read
The authors introduce EDiS, a framework that performs a one-time decomposition of a graph into edge-disjoint subgraphs, then efficiently recombines these subgraphs per epoch to produce training graphs with varying edge budgets.
Why this paper
From Adobe Research and 5 others
In one line
EDiS decomposes a graph once into edge-disjoint subgraphs and recombines them per epoch to produce variable sparse graphs for efficient GNN training.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§