Research map / Label-efficient and robust learning / Noisy labels
Simple and Effective Regularization Methods for Training on Noisily Labeled Data with Generalization Guarantee
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 |
|---|---|---|---|
| Add per-example trainable auxiliary variables | Wide fully-connected network trained with GD · MNIST, 5 vs. 8, varying label noise | test error | Helped |
| Regularize parameter distance to initialization | Wide fully-connected network trained with GD · MNIST, 5 vs. 8, varying label noise | test error | Helped |
| Use AUX or RDI instead of early stopping | Wide fully-connected network trained with GD · MNIST, 5 vs. 8, varying label noise | test error | No clear effect |
| Replace AUX with RDI | Wide fully-connected network trained with GD · MNIST, 5 vs. 8, varying label noise | test error | No clear effect |
| Add per-example trainable auxiliary variables | 11-layer CNN trained with SGD · CIFAR, airplanes vs. automobiles, 20% label noise | test error | Helped |
| Regularize parameter distance to initialization | 11-layer CNN trained with SGD · CIFAR, airplanes vs. automobiles, 20% label noise | test error | Helped |
| Use AUX or RDI instead of early stopping | 11-layer CNN trained with SGD · CIFAR, airplanes vs. automobiles, 20% label noise | test error | No clear effect |
| Replace AUX with RDI | 11-layer CNN trained with SGD · CIFAR, airplanes vs. automobiles, 20% label noise | test error | Mixed |
| Add per-example trainable auxiliary variables | ResNet-34 · CIFAR-10, 10 classes, noise rate 0.4 | test accuracy | Helped |
| Continue AUX training through the last epoch instead of selecting its best epoch | ResNet-34 with AUX · CIFAR-10, 10 classes, noise rate 0.4 | test accuracy | No clear effect |
| Use categorical cross-entropy instead of MSE (ablation) | ResNet-34 with AUX · CIFAR-10, 10 classes, noise rate 0.4 | test accuracy | No clear effect |
| Use the difference trick to make initial output zero while preserving the NTK (ablation) | Neural network initialization · Experiments; setting not specified | performance | Helped |
Closest papers by meaning
- Understanding deep learning requires rethinking generalization (ICLR 2017 · similarity 0.90)
- Regularization Matters: Generalization and Optimization of Neural Nets v.s. their Induced Kernel (NeurIPS 2019 · similarity 0.89)
- Network as Regularization for Training Deep Neural Networks: Framework, Model and Performance (AAAI 2020 · similarity 0.89)
- Fine-Grained Analysis of Optimization and Generalization for Overparameterized Two-Layer Neural Networks (ICML 2019 · similarity 0.89)
- How Does Label Noise Gradient Descent Improve Generalization in the Low SNR Regime? (NeurIPS 2025 · similarity 0.89)
- To Grok or not to Grok: Disentangling Generalization and Memorization on Corrupted Algorithmic Datasets (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.