This repository presents the implementation of the RDS-Contrastive model, proposed to handle noisy samples during the training of machine learning models.
The method focuses on improving robustness under label noise scenarios using a contrastive learning strategy combined with RDS mechanisms.
You can push the Docker image using:
docker push vitorbds/tf2_llp:tagname
docker pull vitorbds/tf2_llp:tagname
python main_coat_rds.py
--noise_type pairflip
--dataset cifar100
--batch_size 128
--noise_rate 0.45
--results /share_alpha_2/vitor/cifar_100_rds/pairflip_45/rodada_5
| Parameter | Description |
|---|---|
--noise_type |
Type of label noise (e.g., pairflip, symmetric) |
--dataset |
Dataset used for training (e.g., cifar100,'cifar10,'mnist') |
--batch_size |
Training batch size |
--noise_rate |
Noise ratio (e.g., 0.45 = 45%) |
--results |
Directory to save experiment results |