Research map / Label-efficient and robust learning / Semi-supervised learning
On Adversarial Mixup Resynthesis
Christopher Beckham, Sina Honari, Vikas Verma, Alex M Lamb, Farnoosh Ghadiri, R Devon Hjelm, Yoshua Bengio, Chris Pal
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 adversarial mixup resynthesis (mixup k=3) to AE+GAN | AE+GAN · SVHN linear probe (d_h=32) | accuracy: 37.01 -> 47.34 | Improved |
| add AMR mixing (mixup k=2) | AE+GAN · MNIST linear probe (d_h=32) | accuracy: 97.52 -> 98.01 | Improved |
| add AMR mixing (Bern k=2) | AE+GAN · KMNIST linear probe (d_h=32) | accuracy: 76.18 -> 81.54 | Improved |
| use mixup k=3 instead of k=2 (ablation) | AMR · SVHN 20k subset (d_h=32) | accuracy: 36.27 -> 37.04 | Improved |
| add AMR mixup k=2 in very low-data regime | AE+GAN · SVHN 1k subset (d_h=32) | accuracy: 22.71 -> 21.89 | No clear effect |
| increase mixup order k from 2 to 8 (ablation) | AMR · SVHN d_h=256 | accuracy: 71.51 -> 75.71 | Improved |
| increase mixup order k from 4 to 8 (ablation) | AMR · CIFAR10 d_h=1024 | accuracy: 61.72 -> 59.75 | Hurt |
| increase mixup order k from 3 to 8 (ablation) | AMR · CIFAR10 d_h=256 | accuracy: 54.94 -> 53.07 | Hurt |
| use Bernoulli mixing instead of mixup (k=2) (ablation) | AMR · SVHN d_h=256 | accuracy: 71.51 -> 58.64 | Hurt |
| use Bernoulli mixing instead of mixup (k=3) (ablation) | AMR · DSprites disentanglement | accuracy: 51.13 -> 56.00 | Improved |
| add AMR mixing (mixup k=2) | AE+GAN · DSprites disentanglement | accuracy: 45.12 -> 49.00 | Improved |
| replace ACAI mixing with AMR mixup k=2 | ACAI (reimplemented) · SVHN d_h=256 | accuracy: 68.64 -> 71.51 | Improved |
Closest papers by meaning
- Adversarial Mixup Unlearning (ICLR 2025 · similarity 0.85)
- Annealing Self-Distillation Rectification Improves Adversarial Training (ICLR 2024 · similarity 0.84)
- Synergy Between Sufficient Changes and Sparse Mixing Procedure for Disentangled Representation Learning (ICLR 2025 · similarity 0.83)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.83)
- MixAT: Combining Continuous and Discrete Adversarial Training for LLMs (NeurIPS 2025 · similarity 0.83)
- DAT: Improving Adversarial Robustness via Generative Amplitude Mix-up in Frequency Domain (NeurIPS 2024 · similarity 0.83)
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.
