Research map / Graph learning / Graph self-supervised learning
Autoencoder-based Graph Construction for Semi-supervised Learning
Mingeun Kang, Kiwon Lee, Yong H. Lee, Changho Suh
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 |
|---|---|---|---|
| add autoencoder-based graph to feature matching (ablation) | supervised + feature matching · CIFAR-10 4000 labels | error rate (%): 13.99 -> 13.37 | Improved |
| add feature matching loss with autoencoder graph (ablation) | Pi-model · CIFAR-10 4000 labels | error rate (%): 12.36 -> 10.81 | Improved |
| use autoencoder graph instead of plain graph (ablation) | Pi-model + feature matching · CIFAR-10 4000 labels | error rate (%): 11.00 -> 10.81 | Improved |
| add feature matching loss with autoencoder graph (ablation) | Mean Teacher · CIFAR-10 4000 labels | error rate (%): 12.31 -> 11.54 | Improved |
| use autoencoder graph instead of plain graph (ablation) | Mean Teacher + feature matching · CIFAR-10 4000 labels | error rate (%): 12.12 -> 11.54 | Improved |
| add feature matching loss with autoencoder graph (ablation) | ICT · CIFAR-10 4000 labels | error rate (%): 7.25 -> 6.87 | Improved |
| use autoencoder graph instead of plain graph (ablation) | ICT + feature matching · CIFAR-10 4000 labels | error rate (%): 7.18 -> 6.87 | Improved |
| add consistency loss to supervised loss (ablation) | supervised only · CIFAR-10 4000 labels | error rate (%): 20.26 -> 12.36 | Improved |
| add feature matching alone to supervised loss (ablation) | supervised only · CIFAR-10 4000 labels | error rate (%): 20.26 -> 13.99 | Improved |
| replace alternating nuclear-norm matrix completion with jointly trained autoencoder | graph-based SSL (GSCNN) · MNIST 100 labels, no augmentation | error rate (%): 62,240.97 s -> 20,108.16 s; 0.60 -> 0.57 | Improved |
| use fewer labels (relative gain over best baseline grows) | Our model · MNIST | error rate (%): 13.6% at 100 labels, 31.9% at 50, 37.6% at 20 | Improved |
| use Wide ResNet-28 instead of CNN-13 with the method | Our model · CIFAR-10 4000 labels | error rate (%): 6.87 -> 6.24 | Improved |
Closest papers by meaning
- Matrix Completion for Graph-Based Deep Semi-Supervised Learning (AAAI 2019 · similarity 0.92)
- A Flexible Generative Framework for Graph-based Semi-supervised Learning (NeurIPS 2019 · similarity 0.89)
- Regularizing Graph Neural Networks via Consistency-Diversity Graph Augmentations (AAAI 2022 · similarity 0.89)
- Data driven semi-supervised learning (NeurIPS 2021 · similarity 0.88)
- The quest for the GRAph Level autoEncoder (GRALE) (NeurIPS 2025 · similarity 0.87)
- Graph Geometry-Preserving Autoencoders (ICML 2024 · 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.
