label noise
The presence of incorrect or uncertain labels in a dataset, which can degrade model performance and reliability, necessitating robust training methods and techniques to mitigate its impact, such as noise-robust algorithms and data cleansing.
- Adam Reduces a Unique Form of Sharpness: Theoretical Insights Near the Minimizer Manifold
- Exploring the Noise Robustness of Online Conformal Prediction
- FlowRefiner: A Robust Traffic Classification Framework against Label Noise
- From Pretraining to Pathology: How Noise Leads to Catastrophic Inheritance in Medical Models
- How Does Label Noise Gradient Descent Improve Generalization in the Low SNR Regime?
- Robust Minimax Boosting with Performance Guarantees
- Towards Robust Parameter-Efficient Fine-Tuning for Federated Learning