Research map / Label-efficient and robust learning / Noisy labels
Bridging Generative and Discriminative Noisy-Label Learning via Direction-Agnostic EM Formulation
Fengbei Liu, Chong Wang, Yuanhong Chen, Yuyuan Liu, Gustavo Carneiro
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 uncertainty component to coverage-only prior (ablation) | PLS prior · CIFAR-10 IDN 50% | accuracy (%): 78.38 -> 90.57 | Improved |
| add coverage component to uncertainty-only prior (ablation) | PLS prior · CIFAR-10 IDN 50% | accuracy (%): 66.61 -> 90.57 | Improved |
| add EMA moving average (beta 0.9) to coverage (ablation) | PLS prior · CIFAR-10 IDN 50% | accuracy (%): 61.47 -> 90.57 | Improved |
| use stochastic sampling instead of argmax for coverage (ablation) | PLS prior · CIFAR-10 IDN 50% | accuracy (%): 15.26 -> 90.57 | Improved |
| use GMM-based noise probability instead of random w_i (ablation) | PLS prior · CIFAR-10 IDN 50% | accuracy (%): 84.28 -> 90.57 | Improved |
| add E-step KL divergence (ablation) | EM objective · CIFAR-10 IDN 50% | accuracy (%): 80.46 -> 90.57 | Improved |
| use L_PRI instead of soft cross-entropy on partial labels (ablation) | EM objective · CIFAR-10 IDN 50% | accuracy (%): 73.72 -> 90.57 | Improved |
| use reverse KL instead of original KL for L_PRI (ablation) | EM objective · CIFAR-10 IDN 50% | accuracy (%): 88.47 -> 90.57 | Improved |
| raise moving-average beta from 0.5 to 0.9 (ablation) | PLS prior · CIFAR-10 IDN 50% | accuracy (%): 85.73 -> 90.57 | Improved |
| increase K from 1 to 3 label samples (ablation) | EM objective · CIFAR-10 IDN 50% | accuracy (%): 90.57 -> 89.75 | No clear effect |
| add full EM with PLS instead of noisy labels only (CE) (ablation) | noisy-label classifier · CIFAR-10 IDN 50% | accuracy (%): 58.93 -> 90.57 | Improved |
| use p(X|Y) instead of p(Y|X) objective | EM objective · CIFAR-10 IDN 50% | accuracy (%): 89.77 -> 90.57 (CIFAR-10 IDN 50%); 54.53 -> 53.34 (Red Mini-ImageNet 20%) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- UNICON — UniCon: Combating Label Noise Through Uniform Selection and Contrastive Learning (CVPR 2022) · 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
- Directional Label Diffusion Model for Learning from Noisy Labels (CVPR 2025 · similarity 0.89)
- Generative Modeling Reinvents Supervised Learning: Label Repurposing with Predictive Consistency Learning (ICML 2025 · similarity 0.89)
- Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · similarity 0.88)
- Imprecise Label Learning: A Unified Framework for Learning with Various Imprecise Label Configurations (NeurIPS 2024 · similarity 0.88)
- L2B: Learning to Bootstrap Robust Models for Combating Label Noise (CVPR 2024 · similarity 0.88)
- Label Noise: Ignorance Is Bliss (NeurIPS 2024 · similarity 0.88)
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.
