Research map / Label-efficient and robust learning / Noisy labels
Latent Class-Conditional Noise Model
Jiangchao Yao, Bo Han, Zhihan Zhou, Ya Zhang, Ivor Wai-Hung Tsang
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 |
|---|---|---|---|
| replace S-adaptation noise-transition layer with latent class-conditional model | noise transition learner · CIFAR-10 symmetric noise r=0.5 | accuracy (%): 83.4 -> 87.3 | Improved |
| use LCCN instead of CE | PreAct ResNet-32 · CIFAR-100 asymmetric noise r=0.4 | accuracy (%): 51.6 -> 65.5 | Improved |
| use LCCN instead of DivideMix | LCCN · CIFAR-10 asymmetric noise r=0.7 | accuracy (%): 75.1 -> 79.6 | Improved |
| add cross-model training and co-refinement tricks (DivideLCCN) | LCCN · CIFAR-10 symmetric noise r=0.5 | accuracy (%): 87.3 -> 95.3 | Improved |
| use DivideLCCN instead of DivideMix | DivideMix · CIFAR-10 symmetric noise r=0.9 | accuracy (%): 75.7 -> 79.2 | Improved |
| use DivideLCCN instead of DivideMix | DivideMix · CIFAR-100 asymmetric noise r=0.4 | accuracy (%): 58.3 -> 73.6 | Improved |
| add open-set noise handling (LCCN*) (ablation) | LCCN · CIFAR-10 open-set noise r=0.7 | accuracy (%): 69.4 -> 72.4 | Improved |
| add a small clean set for calibration (LCCN+) (ablation) | LCCN* · CIFAR-10 open-set noise r=0.7 | accuracy (%): 72.4 -> 87.5 | Improved |
| add clean data to LCCN* (LCCN*+) (ablation) | LCCN+ · CIFAR-10 open-set noise r=0.7 | accuracy (%): 87.5 -> 89.4 | Improved |
| use DivideLCCN instead of LCCN+ at extreme open-set noise | LCCN+ · CIFAR-10 open-set noise r=0.7 | accuracy (%): 87.5 -> 79.2 | Hurt |
| use LCCN instead of Bootstrapping | Bootstrapping · CIFAR-100 asymmetric noise r=0.5 | accuracy (%): 34.5 -> 33.7 | No clear effect |
| use T-revision instead of CE at high symmetric noise | CE · CIFAR-10 symmetric noise r=0.9 | accuracy (%): 42.8 -> 28.6 | Hurt |
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.
- DivideMix — DivideMix: Learning with Noisy Labels as Semi-supervised Learning (ICLR 2020) · better on 5 of 5 shared results
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
- Safeguarded Dynamic Label Regression for Noisy Supervision (AAAI 2019 · similarity 0.96)
- Learning from Noisy Labels with No Change to the Training Process (ICML 2021 · similarity 0.92)
- Beyond Class-Conditional Assumption: A Primary Attempt to Combat Instance-Dependent Label Noise (AAAI 2021 · similarity 0.91)
- Extended : Learning with Mixed Closed-set and Open-set Noisy Labels (TPAMI 2022 · similarity 0.90)
- Learning the Latent Causal Structure for Modeling Label Noise (NeurIPS 2024 · similarity 0.89)
- Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space (AAAI 2025 · 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.
