An implement of the NeurIPS 2022 paper: [Asymptotically Unbiased Instance-wise Regularized Partial AUC Optimization: Theory and Algorithm].

• Ubuntu 16.04
• CUDA 10.2
• Python 3.7.3
• Pytorch 1.12.1+cu10
• Numpy 1.21.4
• pandas 1.3.5
• scikit-learn 1.0.1

Download cifar-10-long-tail, cifar-100-long-tail, and tiny-imagenet-200. Unzip these files and place then in ./data/[dataset]/.

Using the pretrained models in ./pretrained_models_back.

Run the following command for training & validation

python3 train_SPAUCI.py

With augmentations.
train set has 10415 images
class number:  [9511  904]
test set has 2236 images
class number:  [2042  194]
val set has 2233 images
class number:  [2039  194]
--------------------------------------------------
To balance ratio, add 1474 pos imgs (with replace = True)
--------------------------------------------------
--------------------------------------------------
after complementary the ratio, having 11889 images
--------------------------------------------------
epoch:0 val pauc:0.9366215047631786
epoch:1 val pauc:0.9495861038852849
...
epoch:48 val pauc:0.9628886839452894
epoch:49 val pauc:0.9652434893251374
test pauc:0.7721479724874244

The following methods are provided in this repository (see section 3 in our paper):

• OPAUC Losses, An implement of the loss function in section 3.1.
• TPAUC Losses, An implement of the loss function in section 3.2.

See ./losses/SPAUCI.py for usage.

An implement of the training algorithm in section 4.

See ./optimizer/MinMax.py for usage.

If this code is helpful to you, please consider citing our paper:

@inproceedings{shao2022pauci,
  title={Asymptotically Unbiased Instance-wise Regularized Partial AUC Optimization: Theory and Algorithm},
  author={Shao, Huiyang and Xu, Qianqian and Yang, Zhiyong and Bao, Shilong and Huang, Qingming},
  booktitle={Annual Conference on Neural Information Processing Systems},
  year={2022}
}