We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Edge-free but Structure-aware: Prototype-Guided Knowledge Distillation from GNNs to MLPs

Abstract: Distilling high-accuracy Graph Neural Networks~(GNNs) to low-latency multilayer perceptrons~(MLPs) on graph tasks has become a hot research topic. However, MLPs rely exclusively on the node features and fail to capture the graph structural information. Previous methods address this issue by processing graph edges into extra inputs for MLPs, but such graph structures may be unavailable for various scenarios. To this end, we propose a Prototype-Guided Knowledge Distillation~(PGKD) method, which does not require graph edges~(edge-free) yet learns structure-aware MLPs. Specifically, we analyze the graph structural information in GNN teachers, and distill such information from GNNs to MLPs via prototypes in an edge-free setting. Experimental results on popular graph benchmarks demonstrate the effectiveness and robustness of the proposed PGKD.
Comments: 8 pages, 4 figures, 9 tables
Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
Cite as: arXiv:2303.13763 [cs.LG]
  (or arXiv:2303.13763v2 [cs.LG] for this version)

Submission history

From: Taiqiang Wu [view email]
[v1] Fri, 24 Mar 2023 02:28:55 GMT (1492kb,D)
[v2] Mon, 27 Mar 2023 12:06:35 GMT (1492kb,D)

Link back to: arXiv, form interface, contact.