Research map / Label-efficient and robust learning / Noisy labels
Variational Rectification Inference for Learning with Noisy Labels
Haoliang Sun, Qi Wei, Lei Feng, Yupeng Hu, Fan Liu, Hehe Fan, Yilong Yin
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 |
|---|---|---|---|
| use VRI instead of Monte-Carlo approximation with 1 sample (ablation) | rectification meta-learner · CIFAR-10 Flip 40% | test accuracy (%): 88.23 -> 90.20 | Improved |
| use VRI with 1 sample instead of MC with 5 samples (ablation) | rectification meta-learner · CIFAR-10 Flip 40% | test accuracy (%): 89.87 -> 90.20 | Improved |
| add Bayesian prior network and sampling versus non-Bayesian version (ablation) | VRI · CIFAR-10 Uniform 40% | test accuracy (%): 89.27 -> 91.29 | Improved |
| add Bayesian prior network and sampling versus non-Bayesian version (ablation) | VRI · CIFAR-10 Instance-dependent 40% | test accuracy (%): 87.01 -> 90.60 | Improved |
| add Bayesian prior network and sampling versus non-Bayesian version (ablation) | VRI · CIFAR-100 Instance-dependent 40% | test accuracy (%): 64.92 -> 68.17 | Improved |
| use Sigmoid instead of tanh in last meta-net layer (ablation) | VRI · CIFAR-10 Uniform 40% | test accuracy (%): 88.17 -> 91.29 | Improved |
| add Sigmoid in last meta-net layer (ablation) | VRI · CIFAR-10 Uniform 40% | test accuracy (%): 86.90 -> 91.29 | Improved |
| use meta-net hidden layers 1024 and 512 instead of 128 (ablation) | VRI · CIFAR-10 Flip 40% | test accuracy (%): 89.76 -> 91.21 | Improved |
| make meta-net deeper (1024,512,512 layers) (ablation) | VRI · CIFAR-10 Flip 40% | test accuracy (%): 91.21 -> 90.61 | Hurt |
| balance classes of pseudo meta-data (ablation) | VRI without clean meta-data · CIFAR-10 Uniform 40% | test accuracy (%): 80.23 -> 91.24 | Improved |
| balance classes of pseudo meta-data (ablation) | VRI without clean meta-data · ANIMAL-10N | test accuracy (%): 81.4 -> 85.8 | Improved |
| use clean meta-data (standard) instead of pseudo meta-data (ablation) | VRI · CIFAR-10 Flip 40% | test accuracy (%): 91.11 -> 91.93 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PADDLES — PADDLES: Phase-Amplitude Spectrum Disentangled Early Stopping for Learning with Noisy Labels (ICCV 2023) · better on 2 of 2 shared results
- USDNL — USDNL: Uncertainty-Based Single Dropout in Noisy Label Learning (AAAI 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Meta-Amortized Variational Inference and Learning (AAAI 2020 · similarity 0.88)
- Recursive Inference for Variational Autoencoders (NeurIPS 2020 · similarity 0.87)
- VIKING: Deep variational inference with stochastic projections (NeurIPS 2025 · similarity 0.87)
- Amortized Variational Inference for Partial-Label Learning: A Probabilistic Approach to Label Disambiguation (ICML 2026 · similarity 0.86)
- Meta-Learning with Shared Amortized Variational Inference (ICML 2020 · similarity 0.86)
- SoftCVI: Contrastive variational inference with self-generated soft labels (ICLR 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.
