Research map / Label-efficient and robust learning / Domain adaptation and generalization
Unsupervised Adversarial Invariance
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 the split representation, competitive reconstruction, and adversarial disentanglement instead of CAI | Unsupervised Adversarial Invariance · Extended Yale-B | A_y accuracy: 0.89 -> 0.95 | Helped |
| Use the split representation, competitive reconstruction, and adversarial disentanglement instead of CAI | Unsupervised Adversarial Invariance · Chairs | A_z accuracy: 0.69 -> 0.34 | Helped |
| Use unsupervised invariance induction rather than supervised CAI | MNIST-ROT · unseen rotations ±55° and ±65° | A_y accuracy: 0.826 -> 0.856; 0.662 -> 0.696 | Helped |
| Use unsupervised invariance induction rather than supervised CAI | MNIST-ROT · MNIST-ROT | A_z accuracy: 0.384 -> 0.338 | Helped |
| Use unsupervised invariance induction rather than supervised CAI | MNIST-DIL · MNIST-DIL, k=-2, 2, 3, 4 | A_y accuracy: 0.816 -> 0.880; 0.933 -> 0.958; 0.795 -> 0.874; 0.519 -> 0.606 | Helped |
| Use unsupervised invariance induction rather than VFAE | Unsupervised Adversarial Invariance · Amazon Reviews, books→electronics | A_y accuracy: 0.792 -> 0.764 | Hurt |
| Use unsupervised invariance induction rather than VFAE | Unsupervised Adversarial Invariance · Amazon Reviews, books→dvd | A_y accuracy: 0.799 -> 0.820 | Helped |
| Use unsupervised invariance induction rather than supervised invariance induction overall | Unsupervised Adversarial Invariance · Amazon Reviews, 12 source-target tasks | A_y accuracy | Mixed |
| Add adversarial disentanglers (ablation) | MNIST-ROT B1 · MNIST-ROT | A_z accuracy: 0.409 -> 0.338 | Helped |
| Add adversarial disentanglers (ablation) | MNIST-DIL B1 · MNIST-DIL, k=-2 | A_y accuracy: 0.870 -> 0.880 | Helped |
| Replace the proposed split-representation model with a feed-forward classifier (ablation) | MNIST-ROT · MNIST-ROT | A_z accuracy: 0.338 -> 0.586 | Hurt |
| Replace the proposed split-representation model with a feed-forward classifier (ablation) | MNIST-DIL · MNIST-DIL, k=-2 | A_y accuracy: 0.880 -> 0.872 | Hurt |
Closest papers by meaning
- Invariant Representations without Adversarial Training (NeurIPS 2018 · similarity 0.91)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.89)
- Learning Invariances in Neural Networks from Training Data (NeurIPS 2020 · similarity 0.89)
- Invariant Representations through Adversarial Forgetting (AAAI 2020 · similarity 0.89)
- Understanding Invariance via Feedforward Inversion of Discriminatively Trained Classifiers (ICML 2021 · similarity 0.88)
- Unsupervised Representation Learning - an Invariant Risk Minimization Perspective (ICLR 2026 · 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.