Research map / Graph learning / Graph self-supervised learning
Learnable Pooling in Graph Convolutional Networks for Brain Surface Analysis
Karthik Kuduva Gopinath, Christian Desrosiers, Hervé Lombaert
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace global average pooling with learnable pooling (ablation) | graph convolution network · Mindboggle sex classification | accuracy: 60.76 -> 81.33 | Improved |
| replace fixed atlas parcellation pooling with learnable pooling (ablation) | graph convolution network · Mindboggle sex classification | accuracy: 64.59 -> 81.33 | Improved |
| replace unsupervised spectral clustering pooling with learned supervised clusters (ablation) | graph convolution network · Mindboggle sex classification | accuracy: 67.94 -> 81.33 | Improved |
| train on 100-node sub-sampled graphs instead of full graphs (ablation) | learnable pooling · Mindboggle sex classification | accuracy: 84.21 -> 55.02 | Hurt |
| train on 50K-node sub-sampled graphs instead of full graphs (ablation) | learnable pooling · Mindboggle sex classification | accuracy: 84.21 -> 81.33 | Mixed |
| test 50K-node-trained model on full graphs (ablation) | learnable pooling · Mindboggle sex classification | accuracy: 81.33 -> 78.94 | Mixed |
| add spectral node coordinates as input (ablation) | graph convolution network · ADNI NC vs MCI | accuracy: 63.71 -> 70.79 | Improved |
| add spectral node coordinates as input (ablation) | graph convolution network · ADNI NC vs AD | accuracy: 76.00 -> 89.33 | Improved |
| add spectral node coordinates as input (ablation) | graph convolution network · ADNI MCI vs AD | accuracy: 74.03 -> 76.92 | Improved |
| give raw (x,y,z) mesh coordinates instead of spectral coordinates (ablation) | graph convolution network · ADNI classification | accuracy | Hurt |
| replace random forest on cortical features with learnable-pooling GCN | random forest (cortical-based) · ADNI NC vs AD | accuracy: 80 -> 89.33 | Improved |
| replace random forest on cortical features with learnable-pooling GCN | random forest (cortical-based) · ADNI MCI vs AD | accuracy: 65 -> 76.92 | Improved |
Closest papers by meaning
- Rethinking pooling in graph neural networks (NeurIPS 2020 · similarity 0.89)
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.88)
- Bias, Measurement Error, and Double-Dipping: When Can GNN Convolutions Help Brain Connectome Prediction? (NeurIPS 2026 · similarity 0.86)
- Enhancing Graph Classification Robustness with Singular Pooling (NeurIPS 2025 · similarity 0.86)
- SkipPool: Improved Sparse Hierarchical Graph Pooling with Differentiable Exploration (AAAI 2025 · similarity 0.85)
- Neurodegenerative Brain Network Classification via Adaptive Diffusion with Temporal Regularization (ICML 2024 · similarity 0.85)
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.
