Research map / Label-efficient and robust learning / Semi-supervised learning
Temporal Ensembling 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 |
|---|---|---|---|
| Use temporal ensembling with unlabeled training data | SVHN classifier · SVHN, 500 labels | classification error rate: 18.44% -> 7.05% | Helped |
| Enable standard augmentations | Temporal ensembling · SVHN, 500 labels | classification error rate: 7.05% -> 5.12% | Helped |
| Use temporal ensembling with unlabeled training data | CIFAR-10 classifier · CIFAR-10, 4000 labels | classification error rate: 18.63% -> 16.55% | Helped |
| Enable standard augmentations | Temporal ensembling · CIFAR-10, 4000 labels | classification error rate: 16.55% -> 12.16% | Helped |
| Replace Π-model with temporal ensembling (ablation) | Self-ensembling · CIFAR-10, 4000 labels, with standard augmentations | classification error rate: 12.36% -> 12.16% | Helped |
| Use temporal ensembling with unlabeled training data | SVHN classifier · SVHN, 1000 labels | classification error rate: 8.11% -> 5.43% | Helped |
| Enable standard augmentations | Temporal ensembling · SVHN, 1000 labels | classification error rate: 4.42% | Helped |
| Add randomly selected unlabeled Tiny Images inputs | Temporal ensembling · CIFAR-100 | classification error rate: 26.30% -> 23.63% | Helped |
| Restrict extra unlabeled images to categories present in CIFAR-100 (ablation) | Temporal ensembling with Tiny Images · CIFAR-100 | classification accuracy | No clear effect |
| Use consistency-based self-ensembling with all labels | Self-ensembling classifier · Fully labeled classification | classification error rate | Helped |
| Use consistency-based self-ensembling during training | Temporal ensembling · Training with incorrect labels | classification error rate | Helped |
| Slowly ramp up the unsupervised loss weight (ablation) | Π-model · Semi-supervised image classification | classification error rate | Helped |
Closest papers by meaning
- Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results (NeurIPS 2017 · similarity 0.89)
- Debiased Self-Training for Semi-Supervised Learning (NeurIPS 2022 · similarity 0.88)
- SELF: Learning to Filter Noisy Labels with Self-Ensembling (ICLR 2020 · similarity 0.88)
- Weighted Ensemble Self-Supervised Learning (ICLR 2023 · similarity 0.87)
- Smooth Neighbors on Teacher Graphs for Semi-Supervised Learning (CVPR 2018 · similarity 0.87)
- Barely-Supervised Learning: Semi-supervised Learning with Very Few Labeled Images (AAAI 2022 · 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.