Research map / Label-efficient and robust learning / Noisy labels
Peer Loss Functions: Learning from Noisy Labels without Knowing Noise Rates
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 peer loss instead of DMI | two-layer ReLU MLP · Twonorm, p=0.5, e-1=0.1, e+1=0.3 | accuracy: 0.974 -> 0.977 | No clear effect |
| Use peer loss instead of Surrogate | two-layer ReLU MLP · Splice, p=0.5, e-1=0.1, e+1=0.3 | accuracy: 0.878 -> 0.919 | Helped |
| Use peer loss instead of DMI | two-layer ReLU MLP · Diabetes, p!=0.5, e-1=0.1, e+1=0.3 | accuracy: 0.83 -> 0.856 | Helped |
| Use peer loss instead of Symm | two-layer ReLU MLP · German, p!=0.5, e-1=0.1, e+1=0.3 | accuracy: 0.709 -> 0.727 | Helped |
| Use peer loss instead of Surrogate | two-layer ReLU MLP · Waveform, p=0.5, e-1=0.1, e+1=0.3 | accuracy: 0.895 -> 0.89 | No clear effect |
| Use peer loss instead of NN | two-layer ReLU MLP · Image, p=0.5, e-1=0.1, e+1=0.3 | accuracy: 0.909 -> 0.906 | No clear effect |
| Use peer loss instead of cross entropy | ResNet · CIFAR-10, error rate 0.2 | accuracy: 86.67 -> 87.72 | Helped |
| Use peer loss instead of cross entropy | ResNet · CIFAR-10, error rate 0.4 | accuracy: 82.09 -> 83.81 | Helped |
| Add a peer term by subtracting loss on independent peer samples | peer loss · Noisy-label binary classification, in expectation | expected peer loss | Helped |
| Use alpha-weighted peer loss and tune alpha on validation data | peer loss with unequal class prior · Binary classification with p != 0.5 | 0-1 risk | Helped |
Closest papers by meaning
- Loss factorization, weakly supervised learning and label noise robustness (ICML 2016 · similarity 0.91)
- Label Noise: Ignorance Is Bliss (NeurIPS 2024 · similarity 0.89)
- Information-theoretic Limits of Online Classification with Noisy Labels (NeurIPS 2024 · similarity 0.88)
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.88)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · similarity 0.87)
- Robust Loss Functions for Training Decision Trees with Noisy Labels (AAAI 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.