GrapNet introduces a programmable neural graph substrate where architecture edits are first-class operations. It outperforms dense MLPs on Split Fashion-MNIST and CIFAR-10, achieving 63.16% and 3.81% accuracy gains respectively, with statistically significant results.
arxiv
arXiv cs.LG
·
7d ago
·
research
GrapNet: A Programmable Dynamic-Architecture Neural Graph Substrate
from English
Benchmarks
| Benchmark | Model | Score |
|---|---|---|
| SWE-bench Verified | GrapNet+ER | 63.16percent |
| SWE-bench Verified | MLP+ER | 51.08percent |
| SWE-bench Verified | GrapNet | 3.81pts |
| SWE-bench Verified | MLP-256 | — |