Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Adversarial Active Learning
Samarth Sinha, Sayna Ebrahimi, Trevor Darrell
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 a trained VAE (adversarial game) instead of a discriminator on raw image space only (ablation) | VAAL · BDD100K | mIoU | Improved |
| train the VAE adversarially instead of keeping it frozen as an autoencoder (ablation) | VAAL · BDD100K | mIoU | Improved |
| add the discriminator on top of a VAE using Wasserstein distance uncertainty (ablation) | VAAL · BDD100K | mIoU | Improved |
| use VAAL instead of Core-set for selection with 100 classes at 20% or more labeled | active learning · CIFAR100 | labels needed for same accuracy: ~2.5% fewer labels, equal to 1250 labels | Improved |
| use a smaller budget b=5% instead of b=10% (ablation) | VAAL and baselines · CIFAR100 | accuracy: b=10% -> b=5% slightly better | Improved |
| increase fraction of noisy oracle labels from 10% to 30% (ablation) | VAAL · CIFAR100 | accuracy: 10%, 20%, 30% noise; all strategies converge to random sampling | Hurt |
| exclude m=10 or m=20 classes from the initial labeled pool (ablation) | VAAL · CIFAR100 | accuracy | No clear effect |
| swap VGG16 for ResNet18 as task model (ablation) | VAAL · CIFAR100 | accuracy gap vs Core-set | No clear effect |
| use VAAL instead of Core-set for sampling cost | active learning · CIFAR10 sampling time | time (s): 75.33 -> 10.59 | Improved |
| use VAAL instead of Core-set on high-dimensional inputs (688x688) | active learning · Cityscapes and BDD100K | mIoU | Improved |
Closest papers by meaning
- Task-Aware Variational Adversarial Active Learning (CVPR 2021 · similarity 0.90)
- Multi-Classifier Adversarial Optimization for Active Learning (AAAI 2023 · similarity 0.88)
- AutoAL: Automated Active Learning with Differentiable Query Strategy Search (ICML 2025 · similarity 0.86)
- Improved Active Learning via Dependent Leverage Score Sampling (ICLR 2024 · similarity 0.85)
- Active Classification with Few Queries under Misspecification (NeurIPS 2024 · similarity 0.85)
- Debiased Active Learning with Variational Gradient Rectifier (AAAI 2025 · similarity 0.85)
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.
