English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

GNN-to-MLP distillation aims to utilize knowledge distillation (KD) to learncomputationally-efficient multi-layer perceptron (student MLP) on graph data bymimicking the output representations of teacher GNN. Existing methods mainlymake the MLP to mimic the GNN predictions over a few class labels. However, theclass space may not be expressive enough for covering numerous diverse localgraph structures, thus limiting the performance of knowledge transfer from GNNto MLP. To address this issue, we propose to learn a new powerful graphrepresentation space by directly labeling nodes' diverse local structures forGNN-to-MLP distillation. Specifically, we propose a variant of VQ-VAE to learna structure-aware tokenizer on graph data that can encode each node's localsubstructure as a discrete code. The discrete codes constitute a codebook as anew graph representation space that is able to identify different local graphstructures of nodes with the corresponding code indices. Then, based on thelearned codebook, we propose a new distillation target, namely soft codeassignments, to directly transfer the structural knowledge of each node fromGNN to MLP. The resulting framework VQGraph achieves new state-of-the-artperformance on GNN-to-MLP distillation in both transductive and inductivesettings across seven graph datasets. We show that VQGraph with betterperformance infers faster than GNNs by 828x, and also achieves accuracyimprovement over GNNs and stand-alone MLPs by 3.90% and 28.05% on average,respectively. Code: https://github.com/YangLing0818/VQGraph.

VQGraph: Rethinking Graph Representation Space for Bridging GNNs and MLPs