Research map / Image generation / GANs
Adversarial Learning of Portable Student Networks
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 GAN discriminator (teaching assistant) on features on top of distillation (ablation) | student network · MNIST | misclassification error: 0.65% -> 0.48% | Helped |
| replace FitNet feature regression with adversarial feature matching (ablation) | student network · MNIST | misclassification error: 0.51% -> 0.48% | Helped |
| use teaching assistant vs standard back-propagation (ablation) | LeNet++ student · MNIST | accuracy: 97.2% -> 99.1% | Helped |
| adversarial feature matching vs FitNet (ablation) | Student4 CIFAR-10 · CIFAR-10 | accuracy: 91.55% -> 91.68% | Helped |
| adversarial feature matching vs knowledge distillation (ablation) | Student4 CIFAR-100 · CIFAR-100 | accuracy: 63.07% -> 65.11% | Helped |
| adversarial feature matching vs FitNet (ablation) | Student1 CIFAR-10 · CIFAR-10 | accuracy: 89.03% -> 89.45% | Helped |
| adversarial feature matching vs FitNet (ablation) | Student2 CIFAR-10 · CIFAR-10 | accuracy: 91.01% -> 91.17% | Helped |
| deeper, thinner student (19 layers, 2.5M params) vs 5-layer teacher | student network · CIFAR-10 | accuracy: 90.21% -> 91.68% | Helped |
| reduce student size to ~250K params (36x compression) | student network · CIFAR-10 | accuracy: 91.68% -> 89.45% | Hurt |
Closest papers by meaning
- Distilling Portable Generative Adversarial Networks for Image Translation (AAAI 2020 · similarity 0.91)
- On-Line Adaptative Curriculum Learning for GANs (AAAI 2019 · similarity 0.89)
- Teachers Do More Than Teach: Compressing Image-to-Image Models (CVPR 2021 · similarity 0.89)
- Hybrid Data-Free Knowledge Distillation (AAAI 2025 · similarity 0.88)
- Learning Student Networks with Few Data (AAAI 2020 · similarity 0.88)
- PeerAiD: Improving Adversarial Distillation from a Specialized Peer Tutor (CVPR 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.