ICLR 2024 Poster | [Paper] | [Code]

Suqin Yuan, Lei Feng, Tongliang Liu

The Label Wave method enables early stopping by analyzing training dynamics in learning with noisy labels, eliminating the need for separate validation data.

@inproceedings{
yuan2024early,
title={Early Stopping Against Label Noise Without Validation Data},
author={Suqin Yuan and Lei Feng and Tongliang Liu},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024}
}

We implement our methods by PyTorch on NVIDIA RTX 3090&4090 GPU. The key environments is as bellow:

• PyTorch, version = 1.11.0
• CUDA, version = 11.3

You should put the CIFAR datasets in the folder .\cifar-10 and .\cifar-100 when you have downloaded them.

Here is a training example:

python3 main.py

Contact: Suqin Yuan (suqinyuan.cs@gmail.com).