Description: The package includes the Python 3.7 implementation of the Eulac approach [1], which aims to identify the augmented classes (classes unobserved during training but emerging in testing) by exploiting unlabeled data. The package includes two implementations of the Eulac approach, where Eulac-RKHS trains classifiers with the kernel-based hypothesis space and Eulac-DNN is based on deep models.
Reference:
[1] Yu-Jie Zhang, Peng Zhao, Lanjihong Ma, and Zhi-Hua Zhou. An unbiased risk estimator for learning with augmented classes. In: Advances in Neural Information Processing Systems 33 (NeurIPS'20), 2020.
ATTN: This package is free for academic usage. You can run it at your own risk. For other purposes, please contact
Prof. Zhi-Hua Zhou.
ATTN2: This package was developed by
Yu-Jie Zhang,
Lanjihong Ma and
Yong Bai. For any problem concerning the code, please feel free to contact Mr. Zhang.
Requirement: This package is developed with Python 3.7 and the dependencies are specified in requirements.txt.
Download: [
code] (87KB) [
dataset] (270MB)