Research map / Label-efficient and robust learning / Semi-supervised learning
MarginGAN: Adversarial Training in 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 unlabeled examples with pseudo labels (ablation) | MarginGAN classifier · MNIST, 600 labels | error rate (%): 8.21 -> 4.54 | Helped |
| add generated fake examples with inverted cross-entropy (ablation) | MarginGAN classifier · MNIST, 600 labels | error rate (%): 4.54 -> 3.20 | Helped |
| add generated fake examples (ablation) | MarginGAN classifier · MNIST, 600 labels | training time (sec.): 1305.64 -> 367.79 | Helped |
| add unlabeled examples (ablation) | MarginGAN classifier · MNIST, 600 labels | training time (sec.): 408.41 -> 1305.64 | Hurt |
| train on unlabeled examples only after pretraining (no labeled) (ablation) | MarginGAN classifier · MNIST | error rate (%): 9.78 -> 89.53 | Hurt |
| add generated examples to unlabeled-only extreme training (ablation) | MarginGAN classifier · MNIST | error rate (%): 89.53 -> 7.40 | Helped |
| train generated-example margin reduction for longer (ablation) | MarginGAN classifier · MNIST | error rate (%) | Hurt |
| use MarginGAN with Mean Teacher classifier instead of Q model alone | Mean Teacher ResNet · CIFAR-10, 1000 labels | error rate (%): 27.36 -> 10.39 | Helped |
| use MarginGAN with Mean Teacher classifier instead of Q model alone | Mean Teacher ResNet · CIFAR-10, 4000 labels | error rate (%): 13.20 -> 6.44 | Helped |
| use MarginGAN with Mean Teacher classifier instead of Q model alone | Mean Teacher ResNet · SVHN, 500 labels | error rate (%): 6.83 -> 6.07 | Helped |
Closest papers by meaning
- Good Semi-supervised Learning That Requires a Bad GAN (NeurIPS 2017 · similarity 0.92)
- Triple Generative Adversarial Nets (NeurIPS 2017 · similarity 0.90)
- Semi-supervised Learning with GANs: Manifold Invariance with Improved Inference (NeurIPS 2017 · similarity 0.89)
- Enhancing TripleGAN for Semi-Supervised Conditional Instance Synthesis and Classification (CVPR 2019 · similarity 0.89)
- Regularizing Discriminative Capability of CGANs for Semi-Supervised Generative Learning (CVPR 2020 · similarity 0.89)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.88)
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.