Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Positive-Incentive Noise: How Noise Benefits Models
Hongyuan Zhang, Sida Huang, Yubin Guo, Xuelong Li
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 VPN ResNet18 generator, joint training | ResNet34 · Tiny ImageNet | accuracy: 29.73 -> 34.20 | Improved |
| add VPN ResNet18 generator, joint training | ViT · Tiny ImageNet | accuracy: 58.07 -> 61.49 | Improved |
| add VPN ResNet18 generator, joint training | ViT · CIFAR-10 (joint training) | accuracy: 79.91 -> 82.26 | Improved |
| add VPN ResNet18 generator, frozen base model | ResNet18 · CIFAR-10 (frozen base model) | accuracy: 94.77 -> 95.49 | Improved |
| add VPN DNN3 generator, frozen base model | ResNet34 · Fashion-MNIST (frozen base model) | accuracy: 93.84 -> 95.99 | Improved |
| add VPN DNN3 generator, fine-tuning | ResNet18 · ILSVRC2012 | accuracy: 69.76 -> 69.87 | No clear effect |
| add random Gaussian noise on 10% pixels (ablation) | ViT · CIFAR-10 (joint training) | accuracy: 79.91 -> 64.38 | Hurt |
| add random Gaussian noise on 10% pixels (ablation) | ResNet18 · ILSVRC2012 | accuracy: 69.76 -> 50.67 | Hurt |
| add random Gaussian noise on 10% pixels (ablation) | ResNet18 · Tiny ImageNet | accuracy: 28.20 -> 32.18 | Improved |
| add random rotation augmentation (ablation) | ResNet18 · CIFAR-10 (joint training) | accuracy: 94.77 -> 89.87 | Hurt |
| add random translation augmentation (ablation) | ResNet18 · ILSVRC2012 | accuracy: 69.76 -> 59.60 | Hurt |
| add test-time noise from the generator (vs no noise) (ablation) | ResNet18 with DNN3 generator · Tiny ImageNet | accuracy: 34.13 -> 33.81 | No clear effect |
Closest papers by meaning
- Enhance Vision-Language Alignment with Noise (AAAI 2025 · similarity 0.84)
- Sign-Based Optimizers Are Effective Under Heavy-Tailed Noise (NeurIPS 2026 · similarity 0.82)
- Variational Learning is Effective for Large Deep Networks (ICML 2024 · similarity 0.82)
- Pi-DUAL: Using privileged information to distinguish clean from noisy labels (ICML 2024 · similarity 0.82)
- Understanding and Mitigating the Label Noise in Pre-training on Downstream Tasks (ICLR 2024 · similarity 0.82)
- Butterfly Effects of SGD Noise: Error Amplification in Behavior Cloning and Autoregression (ICLR 2024 · similarity 0.82)
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.
