Research map / Label-efficient and robust learning / Semi-supervised learning
Adversarial Dropout for Supervised and 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 virtual adversarial dropout rather than the Pi model | Semi-supervised classifier · MNIST, 1,000 labels | error rate (%): 1.00 $\pm$ 0.08 -> 0.99 $\pm$ 0.07 | Helped |
| Use supervised adversarial dropout rather than the Pi model | Supervised classifier · MNIST, 60,000 labels | error rate (%): 0.50 $\pm$ 0.02 -> \textbf{0.46} $\pm$ \textbf{0.01} | Helped |
| Use VAdD (KL) rather than the strongest listed baseline | Semi-supervised classifier · SVHN, 73,257 labels | error rate (%): 2.53 $\pm$ 0.05 -> \textbf{2.31} $\pm$ \textbf{0.01} | Helped |
| Combine VAdD (KL) with VAT (ablation) | Semi-supervised classifier · SVHN, 1,000 labels | error rate (%): 3.74 $\pm$ 0.09 -> \textbf{3.55} $\pm$ \textbf{0.05} | Helped |
| Combine VAdD (KL) with VAT (ablation) | Semi-supervised classifier · SVHN, 73,257 labels | error rate (%): 2.31 $\pm$ 0.01 -> \textbf{2.23} $\pm$ \textbf{0.03} | Helped |
| Combine VAdD (QE) with VAT (ablation) | Semi-supervised classifier · CIFAR-10, 4,000 labels | error rate (%): 11.32 $\pm$ 0.11 -> \textbf{9.22} $\pm$ \textbf{0.10} | Helped |
| Combine VAdD (KL) with VAT (ablation) | Semi-supervised classifier · CIFAR-10, 50,000 labels | error rate (%): 5.27 $\pm$ 0.10 -> \textbf{4.40} $\pm$ \textbf{0.12} | Helped |
| Combine VAdD (QE) with VAT (ablation) | Semi-supervised classifier · CIFAR-10, 50,000 labels | error rate (%): 5.24 $\pm$ 0.12 -> 4.73 $\pm$ 0.04 | Helped |
| Use VAdD (QE) instead of VAdD (KL) (ablation) | Semi-supervised classifier · CIFAR-10, 4,000 labels | error rate (%): 11.68 $\pm$ 0.19 -> \textbf{11.32} $\pm$ \textbf{0.11} | Helped |
| Use VAdD (QE) instead of VAdD (KL) (ablation) | Semi-supervised classifier · SVHN, 1,000 labels | error rate (%): 3.74 $\pm$ 0.09 -> 4.26 $\pm$ 0.14 | Hurt |
| Use VAdD alone rather than combining it with VAT (ablation) | Semi-supervised classifier · CIFAR-10, 4,000 labels | error rate (%): 11.32 $\pm$ 0.11 -> \textbf{9.22} $\pm$ \textbf{0.10} | Hurt |
| Apply adversarial dropout rather than standard dropout | MNIST autoencoder · MNIST test set | activation sparsity | Helped |
Closest papers by meaning
- Adversarial Dropout for Recurrent Neural Networks (AAAI 2019 · similarity 0.93)
- Dropout Reduces Underfitting (ICML 2023 · similarity 0.90)
- Demystifying Dropout (ICML 2019 · similarity 0.90)
- Dropout with Expectation-linear Regularization (ICLR 2017 · similarity 0.89)
- The Implicit and Explicit Regularization Effects of Dropout (ICML 2020 · similarity 0.88)
- Stochastic Modified Equations and Dynamics of Dropout Algorithm (ICLR 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.