Research map / Label-efficient and robust learning / Noisy labels
Learning from Multiple Annotator Noisy Labels via Sample-Wise Label Fusion
Zhengqi Gao, Fan-Keng Sun, Mingran Yang, Sucheng Ren, Zikai Xiong, Marc Engeler, Antonio Burazer, Linda Wildling, Luca Daniel, Duane S. Boning
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 sample-wise confusion matrices to the weight-vector-only model (ablation) | Ours · CIFAR-100, eps=22 | accuracy: 63.59 -> 64.48 | Improved |
| add sample-wise confusion matrices to the weight-vector-only model (ablation) | Ours · CIFAR-100, eps=23 | accuracy: 59.40 -> 60.12 | Improved |
| add sample-wise weight vectors to the confusion-matrix-only model (ablation) | Ours · CIFAR-100, eps=22 | accuracy: 60.25 -> 64.48 | Improved |
| add sample-wise weight vectors to the confusion-matrix-only model (ablation) | Ours · CIFAR-100, eps=23 | accuracy: 57.07 -> 60.12 | Improved |
| replace majority voting with sample-wise label fusion | Ours vs Mjv · MNIST, eps=30 | accuracy: 87.72 -> 92.49 | Improved |
| replace the second best method with sample-wise label fusion | Ours · MNIST, eps=30 | accuracy: 4.77% more accuracy | Improved |
| replace the second best method with sample-wise label fusion | Ours · CIFAR-100, eps=22 | accuracy: 5.74% more accuracy | Improved |
| replace WDN with sample-wise label fusion | Ours · ImageNet-100, eps=580 | accuracy: 56.03 -> 73.34 | Improved |
| replace MBEM with sample-wise label fusion | Ours · ImageNet-100, eps=680 | accuracy: 20.06 -> 54.93 | Improved |
| replace the best baseline with sample-wise label fusion | Ours · CIFAR-100 with pretrained neural annotators | accuracy: 72.08 -> 73.14 | Improved |
| compare sample-wise label fusion with MBEM when 3 annotators are always correct | Ours · MNIST, N_correct=3 | accuracy: 88.31% vs 88.26% | No clear effect |
| vary lambda of the diagonal regularizer (ablation) | Ours · CIFAR-100, eps=22/23 | accuracy: best at lambda=1.0 | Mixed |
Closest papers by meaning
- Learning From Noisy Labels by Regularized Estimation of Annotator Confusion (CVPR 2019 · similarity 0.90)
- Learning with Instance-Dependent Label Noise: A Sample Sieve Approach (ICLR 2021 · similarity 0.90)
- Noise-Robust Learning from Multiple Unsupervised Sources of Inferred Labels (AAAI 2022 · similarity 0.89)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.88)
- Collaborative Refining for Learning from Inaccurate Labels (NeurIPS 2024 · similarity 0.88)
- Self-cognitive Denoising in the Presence of Multiple Noisy Label Sources (ICML 2024 · similarity 0.87)
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.
