Research map / Label-efficient and robust learning / Noisy labels
Estimating Instance-dependent Bayes-label Transition Matrix using a Deep Neural Network
Shuo Yang, Erkun Yang, Bo Han, Yang Liu, Min Xu, Gang Niu, Tongliang Liu
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 BLTM-V instead of PTD | noisy-label classifier · CIFAR-10 IDN-10% | classification accuracy: 76.33 ± 0.38 -> 82.16 ± 1.01 | Improved |
| Use BLTM-V instead of PTD | noisy-label classifier · CIFAR-10 IDN-30% | classification accuracy: 75.42 ± 1.33 -> 78.82 ± 1.07 | Improved |
| Use BLTM-V instead of PTD | noisy-label classifier · CIFAR-10 IDN-50% | classification accuracy: 56.63 ± 1.88 -> 60.33 ± 5.29 | Improved |
| Use BLTM-V instead of PTD | noisy-label classifier · SVHN IDN-10% | classification accuracy: 93.77 ± 0.33 -> 96.37 ± 0.77 | Improved |
| Use BLTM-V instead of PTD | noisy-label classifier · SVHN IDN-30% | classification accuracy: 89.64 ± 1.98 -> 94.69 ± 0.24 | Improved |
| Use BLTM-V instead of PTD | noisy-label classifier · SVHN IDN-50% | classification accuracy: 71.57 ± 3.32 -> 78.71 ± 4.37 | Improved |
| Use BLTM-V instead of PTD | noisy-label classifier · F-MNIST IDN-30% | classification accuracy: 87.86 ± 0.78 -> 92.24 ± 1.87 | Improved |
| Use BLTM-V instead of PTD | noisy-label classifier · F-MNIST IDN-50% | classification accuracy: 73.38 ± 2.25 -> 76.89 ± 4.26 | Improved |
| Use BLTM-V instead of PTD | noisy-label classifier · Clothing1M | classification accuracy: 70.07 -> 73.39 | Improved |
| Add transition matrix revision (-V) (ablation) | BLTM on F-MNIST IDN noise · F-MNIST IDN-10% to IDN-50% | classification accuracy: 96.06 ± 0.71 -> 96.93 ± 0.31; 94.97 ± 0.33 -> 95.55 ± 0.59; 91.47 ± 1.36 -> 92.24 ± 1.87; 82.88 ± 2.72 -> 83.43 ± 1.72; | Improved |
| Add transition matrix revision (-V) (ablation) | BLTM on CIFAR-10 IDN noise · CIFAR-10 IDN-10% to IDN-50% | classification accuracy: 81.73 ± 0.56 -> 82.16 ± 1.01; 80.26 ± 0.63 -> 80.37 ± 1.98; 77.69 ± 1.37 -> 78.82 ± 1.07; 71.96 ± 2.27 -> 72.93 ± 4.00; | Improved |
| Add transition matrix revision (-V) (ablation) | BLTM on SVHN IDN noise · SVHN IDN-10% to IDN-50% | classification accuracy: 96.05 ± 0.32 -> 96.37 ± 0.77; 94.97 ± 0.58 -> 95.12 ± 0.40; 93.99 ± 1.24 -> 94.69 ± 0.24; 87.67 ± 1.29 -> 88.13 ± 3.23; | Improved |
Closest papers by meaning
- A Parametrical Model for Instance-Dependent Label Noise (TPAMI 2023 · similarity 0.98)
- Estimation of the Label-Noise Transition Matrix with Performance Guarantees via Selective Classification (NeurIPS 2026 · similarity 0.91)
- Learning the Latent Causal Structure for Modeling Label Noise (NeurIPS 2024 · similarity 0.89)
- Learning Causal Transition Matrix for Instance-dependent Label Noise (AAAI 2025 · similarity 0.88)
- Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 2024 · similarity 0.88)
- Dirichlet-based Per-Sample Weighting by Transition Matrix for Noisy Label Learning (ICLR 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.
