Research map / Graph learning / Graph self-supervised learning
Representation Learning on Graphs with Jumping Knowledge Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| add JK layer aggregation with max-pooling | GCN · Citeseer | accuracy: 77.3 (1.3) -> 77.7 (0.5) | Helped |
| add JK layer aggregation with concatenation | GCN · Citeseer | accuracy: 77.3 (1.3) -> 78.3 (0.8) | Helped |
| add JK layer aggregation with max-pooling | GCN · Cora | accuracy: 88.2 (0.7) -> 89.6 (0.5) | Helped |
| add JK layer aggregation with concatenation | GCN · Cora | accuracy: 88.2 (0.7) -> 89.1 (1.1) | Helped |
| use LSTM-attention instead of max-pooling or concatenation for layer aggregation (ablation) | GCN-based JK-Net · Citeseer and Cora | accuracy: 77.7 (0.5) -> 74.7 (0.9); 89.6 (0.5) -> 85.8 (1.0) | Hurt |
| add JK layer aggregation to GraphSAGE using mean node aggregation and concatenation | GraphSAGE · Reddit | Micro-F1: 0.950 -> 0.953 | Helped |
| use LSTM rather than concatenation for layer aggregation with mean node aggregation (ablation) | GraphSAGE-based JK-Net · Reddit | Micro-F1: 0.953 -> 0.955 | Helped |
| use max-pooling rather than concatenation for layer aggregation with max-pooling node aggregation (ablation) | GraphSAGE-based JK-Net · Reddit | Micro-F1: 0.924 -> 0.948 | Helped |
| use LSTM rather than concatenation for layer aggregation with max-pooling node aggregation (ablation) | GraphSAGE-based JK-Net · Reddit | Micro-F1: 0.924 -> 0.877 | Hurt |
| add LSTM-attention layer aggregation to GraphSAGE with mean node aggregation | GraphSAGE · PPI, 30 epochs | Micro-F1: 0.690 -> 0.818 | Helped |
| replace concatenation with LSTM-attention layer aggregation (ablation) | GAT-based JK-Net · PPI | Micro-F1: 0.959 (0.003) -> 0.969 (0.006) | Helped |
| add JK LSTM-attention aggregation to GAT | GAT · PPI | Micro-F1: 0.968 (0.002) -> 0.969 (0.006) | No clear effect |
Closest papers by meaning
- Deep Neural Networks for Learning Graph Representations (AAAI 2016 · similarity 0.89)
- Semi-Supervised Classification with Graph Convolutional Networks (ICLR 2017 · similarity 0.88)
- Convolutional 2D Knowledge Graph Embeddings (AAAI 2018 · similarity 0.87)
- Deep Gaussian Embedding of Graphs: Unsupervised Inductive Learning via Ranking (ICLR 2018 · similarity 0.87)
- Inductive Representation Learning on Large Graphs (NeurIPS 2017 · similarity 0.87)
- Rethinking Kernel Methods for Node Representation Learning on Graphs (NeurIPS 2019 · similarity 0.87)
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.