• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>International journal of innovative computing, information and control >ACWGAN: AN AUXILIARY CLASSIFIER WASSERSTEIN GAN-BASED OVERSAMPLING APPROACH FOR MULTI-CLASS IMBALANCED LEARNING
【24h】

ACWGAN: AN AUXILIARY CLASSIFIER WASSERSTEIN GAN-BASED OVERSAMPLING APPROACH FOR MULTI-CLASS IMBALANCED LEARNING

机译：ACWGAN: AN AUXILIARY CLASSIFIER WASSERSTEIN GAN-BASED OVERSAMPLING APPROACH FOR MULTI-CLASS IMBALANCED LEARNING

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相关主题

摘要

Learning from multi-class imbalance data is a common but challenging task in machine learning community. Oversampling method based on Generative Adversarial Networks (GAN) is an effective countermeasure. However, due to the scarce number of trainable minority samples, existing methods may produce noise or low-quality minority samples; besides, they may suffer from mode collapse. To address the issues, we propose an Auxiliary Classifier Was serstein Generative Adversarial Networks (ACWGAN) for imbalanced dataset. An independent auxiliary classifier is introduced to help discriminator determine whether the minority samples match the corresponding labels, more importantly, to improve the quality of generated minority samples. Furthermore, we use Wasserstein distance instead of Jensen-Shannon divergence in ACWGAN as the distance measure of the probability distribution to alleviate the mode collapse. Extensive experimental testing is performed on 16 multi-class imbalanced benchmarks and two real imbalanced datasets in comparison with several popular oversampling approaches. The experiment result demonstrates that our method is superior to other oversampling approaches.

著录项

获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号