Research map / Label-efficient and robust learning / Semi-supervised learning
Smooth Neighbors on Teacher Graphs for Semi-Supervised Learning
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 SNTG to Pi model | Pi model · MNIST, 20 labels, no augmentation | error rate (%): 6.32$\pm$6.90* -> 1.36$\pm$0.78 | Helped |
| Add SNTG to Pi model | Pi model · CIFAR-10, 4000 labels, no augmentation | error rate (%): 16.55$\pm$0.29 -> 13.62$\pm$0.17 | Helped |
| Add SNTG to VAT+Ent | VAT+Ent · CIFAR-10, 4000 labels, no augmentation | error rate (%): 13.15 -> 12.49$\pm$0.36 | Helped |
| Add SNTG to Pi model | Pi model · SVHN, 250 labels, translation augmentation | error rate (%): 9.93$\pm$1.15* -> 5.07$\pm$0.25 | Helped |
| Add SNTG to Mean Teacher | Mean Teacher · SVHN, 500 labels, translation augmentation | error rate (%): 4.18$\pm$0.27 -> 3.99$\pm$0.24 | Helped |
| Add SNTG to TempEns | TempEns · CIFAR-10, 1000 labels, standard augmentation | error rate (%): 23.31$\pm$1.01* -> 18.41$\pm$0.52 | Helped |
| Remove neighbor-graph regularization R_S (ablation) | Pi+SNTG · CIFAR-10, 4000 labels, no augmentation | error rate (%): 13.62 -> 16.55 | Hurt |
| Add neighbor-graph regularization R_S without consistency regularization R_C (ablation) | Supervised loss · CIFAR-10, 4000 labels, no augmentation | error rate (%): 35.56 -> 15.36 | Helped |
| Remove consistency regularization R_C while keeping R_S (ablation) | Pi+SNTG · CIFAR-10, 4000 labels, no augmentation | error rate (%): 13.62 -> 15.36 | Hurt |
| Replace the learned teacher graph with a fixed input-space k-NN graph (ablation) | Pi+SNTG · CIFAR-10, 4000 labels, no augmentation | error rate (%): 13.62 -> 16.13\% | Hurt |
| Replace the jointly learned teacher graph with a fixed graph from pretrained Pi predictions (ablation) | Pi+SNTG · CIFAR-10, 4000 labels, no augmentation | error rate (%): 13.62 -> 15.71\% | Hurt |
| Add SNTG to FM GAN | FM GAN · CIFAR-10, 4000 labels | error rate (%): 18.63\% -> 14.93\% | Helped |
Closest papers by meaning
- A Flexible Generative Framework for Graph-based Semi-supervised Learning (NeurIPS 2019 · similarity 0.87)
- Temporal Ensembling for Semi-Supervised Learning (ICLR 2017 · similarity 0.87)
- Graph Random Neural Networks for Semi-Supervised Learning on Graphs (NeurIPS 2020 · similarity 0.87)
- Neighborhood-Regularized Self-Training for Learning with Few Labels (AAAI 2023 · similarity 0.87)
- Revisiting Semi-Supervised Learning with Graph Embeddings (ICML 2016 · similarity 0.87)
- Unsupervised Neighborhood Propagation Kernel Layers for Semi-supervised Node Classification (AAAI 2024 · similarity 0.86)
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.