AI Paper Atlas

Research map / Graph learning / GNN architectures

Interpretable Prototype-based Graph Information Bottleneck

Sangwoo Seo, Sungwon Kim, Chanyoung Park

NeurIPS 2023 · Graph learning › GNN architectures · approach: GNN explainability · 8 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
merge similar prototypes (ablation)PGIB_cont · ZINC RLMFidelity F-: 0.441 -> 0.415Improved
merge similar prototypes (ablation)PGIB_cont · ZINC HLM-CLintFidelity F-: 0.593 -> 0.543Improved
merge similar prototypes (ablation)PGIB_cont · ZINC HLM-CLintFidelity F+: 0.772 -> 0.796Improved
merge similar prototypes (ablation)PGIB_cont · ZINC DRD2Fidelity F+: 0.771 -> 0.781Improved
contrastive objective instead of variational IBPGIB · MUTAGaccuracy: 85.00 -> 85.50Improved
contrastive objective instead of variational IBPGIB · PROTEINSaccuracy: 77.14 -> 77.50Improved
mean-pooling readout instead of max-pooling (ablation)PGIB_cont · MUTAGaccuracy: 85.50 -> 88.50Improved
mean-pooling readout instead of max-pooling (ablation)PGIB_cont · PROTEINSaccuracy: 77.50 -> 68.39Hurt
sum-pooling readout instead of max-pooling (ablation)PGIB_cont · NCI1accuracy: 78.25 -> 61.82Hurt
PGIB_cont instead of VGIBVGIB · NCI1accuracy: 63.75 -> 78.25Improved
remove mutual information terms I(G;Gsub) or I(Gsub;Gp) (ablation)PGIB · graph classificationaccuracyHurt
remove prototype merging (ablation)PGIB and PGIB_cont · graph classificationaccuracyHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy