# P-DIFF: Learning Classifier with Noisy Labels based on Probability Difference Distributions

arXiv: https://arxiv.org/abs/2009.06382 · HTML: https://mlsanity.com/paper/2009.06382/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 14 Sep 2020.

## Built on by (most cited first, 7 of 7)

- [In-Context Meta LoRA Generation](https://mlsanity.com/paper/2501.17635/) (2025)
- [Data Efficacy for Language Model Training](https://mlsanity.com/paper/2506.21545/) (2025)
- [Label Noise-Robust Learning using a Confidence-Based Sieving Strategy](https://mlsanity.com/paper/2210.05330/) (2022)
- [LNL+K: Enhancing Learning with Noisy Labels Through Noise Source Knowledge Integration](https://mlsanity.com/paper/2306.11911/) (2023)
- [Noise-Aware Generalization: Robustness to In-Domain Noise and Out-of-Domain Generalization](https://mlsanity.com/paper/2504.02996/) (2025)
- [Semantic-guided LoRA Parameters Generation](https://mlsanity.com/paper/2509.10535/) (2025)
- [A Unified Framework for Connecting Noise Modeling to Boost Noise Detection](https://mlsanity.com/paper/2312.00827/) (2023)

## Abstract (arXiv)

Learning deep neural network (DNN) classifier with noisy labels is a challenging task because the DNN can easily over-fit on these noisy labels due to its high capability. In this paper, we present a very simple but effective training paradigm called P-DIFF, which can train DNN classifiers but obviously alleviate the adverse impact of noisy labels. Our proposed probability difference distribution implicitly reflects the probability of a training sample to be clean, then this probability is employed to re-weight the corresponding sample during the training process. P-DIFF can also achieve good performance even without prior knowledge on the noise rate of training samples. Experiments on benchmark datasets also demonstrate that P-DIFF is superior to the state-of-the-art sample selection methods.